This PDF file contains the front matter associated with SPIE Proceedings Volume 10751, including the Title Page, Copyright information, Table of Contents, and Conference Committee listing.

Stereo matching is challenging due to the presence of perspective distortions and noise. Commonly, stereo matching algorithms utilize local matching techniques to determine the correspondence of two pixels of the same point in a scene. This work presents a stereo matching algorithm based on locally-adaptive windows and correlation filtering. The proposed algorithm estimates the disparity of each pixel by matching an adaptive sliding-window obtained from the left image in the right image of the stereo pair. Computer simulation results obtained with the proposed algorithm are analyzed and discussed by processing pairs of stereo images.

GeoVisipedia (Geospatial Visual Wikipedia) is a new and novel approach to sharing knowledge about complex geospatial entities such as facilities. Facilities are composed of interconnected objects such as buildings, chemical processing units, electrical generation equipment and similar structures. Satellite imagery of a facility reveals a great deal about the organization and visual appearance of objects in a facility, but very little about the identity or function of the object. For example, given a satellite imagery of an oil refinery, an expert in refining readily identifies distillation units and can explain how they work. A non-expert would have a very difficult time identifying these objects let alone explaining how they function. To make highly complex information accessible to non-experts, GeoVisipedia associates a wiki page with objects in satellite imagery. A user selects an object in the image and a wiki page appears that provides the user with detailed information about the object. Experts can author information into the wiki and this information is shared with other users. Additionally, GeoVisipedia automatically transfers all wiki pages from one image of a facility to other imagery of the facility. Consequentially, knowledge about objects in the facility integrates over time as new imagery becomes available and as new wiki pages are created and additional information is added to existing wiki pages. In this respect, satellite imagery becomes a portal to expert knowledge and insight about objects in a facility.

We present a fast and simple technique to generate phase only Fresnel holograms of 2D intensity objects. This technique uses a modified Gerchberg-Saxton (G-S) algorithm to optimize a random phase. This optimization takes into account the resolution, pixel size, reconstruction plane, and optical characteristics of the system that will reconstruct the holograms. The resulting Optimized Fresnel RAndom Phase (OFRAP) is then multiplied with the desired intensity target, and after performing a Fresnel transform and a phase extraction we obtain the corresponding phase only hologram. Numerical results show that generation of holograms with OFRAP achieves performance close to the traditional G-S algorithm directly applied to the intensity target. The proposed technique has the additional advantage that a single ORAP can be used to generate any number of holograms, thus eliminating the need for any further iterative algorithms. This proposal is ideal for the generation of holographic videos and other applications where dynamic generation of phase holograms to manipulate the light field are necessary, like neuronal photostimulation, holographic displays and aberration correction.

Liquid crystal on silicon (LCoS) displays have become the most attractive microdisplays for all sort of spatial light modulation applications. Among the different LCoS technologies, parallel aligned LCoS (PA-LCoS) are especially interesting. They offer unique capabilities as spatial light modulators (SLM), since they enable phase-only modulation without coupled amplitude modulation, and with millions of addressable pixels. In this work we evaluate and demonstrate the various capabilities offered by a novel and simplified physical model for parallel aligned liquid crystal devices (PA-LC), thus also applicable to PA-LCoS microdisplays. The model provides the voltage dependent retardance values, necessary for spatial light modulation applications, for a very wide range of incidence angles and any wavelength in the visible. First, it needs to be calibrated through a reverse-engineering approach. In this calibration, the values obtained for two of the three parameters provide the correct values for important internal properties of these devices related with the birefringence, cell gap and director profile. Therefore, the proposed model can be used as a means to inspect internal physical properties of the cell. Last but not least the model is useful to simulate the retardance for novel PA-LC devices as a function of the LC compound and cell gap. Therefore, it is not only a reverse-engineering model but it also constitutes an analytical alternative to the usual numerical approaches for PA-LC devices design and construction.

A single-pixel camera composed of optical coding masks, a photo detector, and a decoding computation has an important feature of no image sensor for imaging, therefore it has very simple optical and electrical architectures. The optical coding masks whose implementation is a novelty of our research are composed of holes on a substrate, and arranged on the circumference of a disk enables us to switch with a rotation of the disk. The main features are small wavelength dependence other than air for wideband spectral imaging, and the simple structure. The spectral imaging for a sample composed of three color films is demonstrated, and the accuracy of the proposed system is evaluated.

Advancements in machine learning (ML) and deep learning (DL) have enabled imaging systems to perform complex classification tasks, opening numerous problem domains to solutions driven by high quality imagers coupled with algorithmic elements. However, current ML and DL methods for target classification typically rely upon algorithms applied to data measured by traditional imagers. This design paradigm fails to enable the ML and DL algorithms to influence the sensing device itself, and treats the optimization of the sensor and algorithm as separate sequential elements. Additionally, this current paradigm narrowly investigates traditional images, and therefore traditional imaging hardware, as the primary means of data collection. We investigate alternative architectures for computational imaging systems optimized for specific classification tasks, such as digit classification. This involves a holistic approach to the design of the system from the imaging hardware to algorithms. Techniques to find optimal compressive representations of training data are discussed, and most-useful object-space information is evaluated. Methods to translate task-specific compressed data representations into non-traditional computational imaging hardware are described, followed by simulations of such imaging devices coupled with algorithmic classification using ML and DL techniques. Our approach allows for inexpensive, efficient sensing systems. Reduced storage and bandwidth are achievable as well since data representations are compressed measurements which is especially important for high data volume systems.

One of today’s megatrends in the industrial environment is additive manufacturing. Faster prototyping, customized products like hearing devices, integrated functions like heatsinks and many other opportunities are offered by this technological development. The opportunity of using different materials and build up 3-D structures is virtually infinite. Another one is the digitalization of almost any product to build up a smart world. This trend leads to a tremendously rising amount of data to be transferred from one place to another. If a wireless transmission is not possible and if the distance is over 100 m glass fiber is the fastest and most secure way for these requirements. In case of most short-range applications up to 100 m primary copper cables or circuit paths are in use because the electrical data transfer is well known. The limited bandwidth of copper asks for new inventions to meet the demands of tomorrow. Regarding both megatrends, the solution for this upcoming bottleneck could be 3-D printed photonic packages. This paper shows a new and innovative way for the customized fabricating of short-range data transmission networks. By Aerosol Jet Printing (AJP) the so called polymer optical waveguides (POW), it is possible to build up 3-D printed light guiding structures with low attenuation on almost any three-dimensional surface. The main advantages of the here presented research are high flexibility, low weight and low costs. After three years of intensive studies the most important key facts (machine settings, geometry, performance) are summarized in this publication.

Convolutional neural networks have become an essential element of spatial deep learning systems. In the prevailing architecture, the convolution operation is performed with Fast Fourier Transforms (FFT) electronically in GPUs. The parallelism of GPUs provides an efficiency over CPUs, however both approaches being electronic are bound by the speed and power limits of the interconnect delay inside the circuits. Here we present a silicon photonics based architecture for convolutional neural networks that harnesses the phase property of light to perform FFTs efficiently. Our all-optical FFT is based on nested Mach-Zender Interferometers, directional couplers, and phase shifters, with backend electro-optic modulators for sampling. The FFT delay depends only on the propagation delay of the optical signal through the silicon photonics structures. Designing and analyzing the performance of a convolutional neural network deployed with our on-chip optical FFT, we find dramatic improvements by up to 102 when compared to state-of-the-art GPUs when exploring a compounded figure-of-merit given by power per convolution over area. At a high level, this performance is enabled by mapping the desired mathematical function, an FFT, synergistically onto hardware, in this case optical delay interferometers.

In this work, we present an analytical methodology to evaluate the response of a three-beam path Mach-Zehnder interferometer (MZI) based on a multiresolution signal analysis using a discrete Wavelet transform (DWT). Here, the MZI was implemented by fusion splicing of a segment of ytterbium-doped double cladding photonic crystal fiber (Yb-d-DCPCF) between two pieces of single mode fibers. Moreover, Yb-d-DCPCF core and claddings were used as the arms of the MZI, while the hole collapsed regions acted as optical fiber couplers. So, the MZI interference pattern occurs of two main modal interferences, between the core and cladding. Hence, as a result, an interference spectrum with aperiodic sinusoidal waveform was observed. In this way, in our proposed methodology we decompose the interference spectrum into eight bands of spatial frequency (SF) using the DWT with a wavelet mother Daubechies order twenty (db20) for detecting the two main modal interferences. The spectral analysis of the decomposed signal reveals that the level 7 and 8 contain the two main modal interference between the core and the cladding 1 and cladding 2, respectively. Additionally, the fundamental mode is obtained in the level 8. In the same way, the decomposed signal provides information about the wavelength and spatial frequency simultaneously that can be used for knowing the energy spectral density. Finally, we claim that the proposed methodology provides the best accuracy in SF of MZI intermodal and DWT indicates the wavelength region in which the propagation modes occur, improving the description of the refractive index difference.

Residue number system (RNS) enables dimensionality reduction of an arithmetic problem by representing a large number as a set of smaller integers, where the number is decomposed by prime number factorization. These reduced problem sets can then be processed in- dependently and in parallel, thus improving computational efficiency and speed. Here we show an optical RNS hardware representation based on integrated nanophotonics. The digit-wise shifting in RNS arithmetic is expressed as spatial routing of an optical signal in 2×2 hybrid photonic-plasmonic switches. Here the residue is represented by spatially shifting the input waveguides relative to the routers’ outputs, where the moduli are represented by the number of waveguides. By cascading the photonic 2×2 switches, we design a photonic RNS adder and a multiplier forming an all-to- all sparse directional network. The advantage of this photonic arithmetic processor is the short (10’s ps) computational execution time given by the optical propagation delay through the integrated nanophotonic router. Furthermore, we show how photonic processing in- the-network leverages the natural parallelism of optics such as wavelength-division-multiplexing in this RNS processor. A key application for such a photonic RNS engine is the functional analysis of convolutional neural networks.

A hybrid optical integrator is a recirculating loop that performs oversampling typically for an analog input using the crossgain modulation (XGM) in a semiconductor optical amplifier (SOA). The modulated input signal changes the gain of the loop through XGM and thus modifies the loop accumulation. This paper presents hybrid optical integrator for an all-optical analog-to-digital converter based on silicon photonics platform. The device consists of silicon waveguides of dimension 220 × 500 nm (thick × width) and approximately 2,800 μm total loop length, input and output grating couplers for 1550 nm signal, directional couplers, and external components (SOA, optical isolator and band-pass filter). All devices are designed for fabrication on the SOI wafer using E-Beam lithography. A theoretical model for the system is developed and simulated with Lumerical software and Matlab that accounts for critical design parameters such as the loop length, coupling coefficient, and gain in the SOA. The 3-dB couplers were designed and loop length was minimized to reduce the propagation loss through the silicon waveguide and increase the sampling frequency. From the simulation, total loss of silicon photonics device at one round trip is approximately 27 dB. The system is characterized for square waves at Giga- Hertz input frequencies. Simulation results show excellent agreement with the theoretical model of leaky integrator. The operation frequency of the integrator can be increased from MHz up to GHz by replacing fiber cable to silicon waveguide, leading to a free-spectral range of approximately 4.17 GHz.

Optical transceivers often require multiple corner case test conditions in order to meet multiple source agreement (MSA) specs. Typically, some tests need to be applied under multiple temperatures using temperature controller, resulting in extensive test time and high manufacturing test cost. In this paper, we introduce machine learning based test methodology for silicon photonics transceiver manufacturing test with large percentage (>90%) of test time reduction. In order to reduce test time, the desire is to test at one temperature corner and apply machine learning techniques to eliminate other temperature corners. We complied wide range of data set from various prerequisite tests and target test data at temperature No.1 as input data set. Target test at temperature No.2 is employed for supervised learning prediction. For production implementation simplicity, we used linear regression model with Tikhonov regularization [1] and reached R2>0.97 of predicted value correlation with physical measurement value.

This work proposes estimating sky radiance distribution curves between 350-2500nm from images captured with a hemispherical digital camera. A novel hardware system simultaneously captured spectral, spatial, and temporal information to acquire accurate physical measurements of the solar/skydome radiance variation. To achieve this goal, we use a custom-built spectral radiance measurement scanner to measure the directional spectral radiance, a pyranometer to measure the irradiance of the entire hemisphere, and a commercial digital camera to capture high-dynamic range (HDR) hemispherical imagery of the sky. We use the measurements obtained from a commercially available digital camera and correlating spectroradiometer measurements to train machine learning (ML) models to estimate whole sky full-spectrum radiance distributions (VIS, VNIR, and SWIR) from a low dimensional RGB input. We train clear, cloudy, and mixed sky models, and cross-validate the estimated radiance distributions with ground-truth data. We highlight important measured and engineered ML features, and we present useful feature engineering techniques employed to minimize model estimation error. Additional contributions of this work include the code for all ML models and experiments, a dataset of all-sky HDR captures with correlating spectroradiometer measurements captured 453 times over 16 days, and an open-source, crossplatform, interactive viewer used to visualize photometric and radiometric data side by side.

In digital photoelasticity, evaluating the stress map is often affected in regions with critical values. This phenomenon is associated to color degradation effect and high fringe densities. It is a consequence of different experimental conditions, such as: type of birefringent material, relative spectral content of light source, relative spectral response of camera sensor, polarization optical elements, load application, etc. In this study field, the main goal accounts for evaluating the stress values, as better as possible, from photoelasticity images. Which turns the view towards the process that allow to acquire photoelasticity images with more complete information. This makes necessary to analyze the possible effects that each element could introduce into the photoelasticity image generation. This paper presents a computational analysis on the effect that different industrial light sources introduces for recovering the stress maps. Hence, four common industrial light sources are considered for generating the photoelasticity images. In this case, results reveal that there are light sources which represent stronger limitations for evaluating the stress, and that Such effect varies with the load increments. This approach is useful for predicting the possible effect that a light source selection could introduce into the stress evaluation process.

Rectangular shapes in observed scenes are transformed to quadrilaterals in resulting images due to perspective distortion. If the aspect ratio of the original rectangle is known, the associated homography can be computed directly and used for perspective correction. However, in camera document scanning applications, the aspect ratio of the imaged rectangle is unknown. In this work, a homography estimation method appropriate for document scanning applications is given. This method does not require a priori knowledge of the aspect ratio of the imaged rectangle nor a calibrated camera. The computational implementation is evaluated experimentally.

Fingerprint identification is one of the richest tools available for the forensic examiner in a crime investigation, mainly because of its intrinsic characteristics of unicity and immutability. Among the various types of fingerprint found in crime scenes, latent fingerprints (that requires chemical and physical processes for their revelation) are the most used, given it’s high incidence in crime scenes and the existence of databases and automated computational systems for processing. However, the methods and techniques currently used as standards for revealing and collecting these latent vestiges ends up in the destruction of the original evidence, also requiring specific skills and materials, which has high cost and difficulty of access to many Forensic Institutes, like the Brazilian IGP. In addition, there is a negative influence of the physical (moisture) and chemical (organic and inorganic substances) characteristics of the surfaces during the revelation of latent fingerprints. In the current research, a prototype adaptable to a smartphone was developed, using a special light configuration to emphasizes the latent fingerprints. The system was able to obtain latent fingerprints on smooth surfaces in a practical, accessible and non-destructible way. However, the need for specific hardware equipped with other light options for different surfaces was observed. Results shows influence of different wavelengths and the positioning of the illumination on indoor and outdoor environments. As the research advances, we expect the system to become more solid, with the outcome of results more adequate to criminal investigation.

Since the polarization image contains abundant information of the object, the detecting ability of imaging system would be improved via observing the polarization features of this object. On the basis of analysis to the polarization characteristics, the active polarized imaging method is introduced in this paper, and active polarized imaging platform is built. Through this system, 3 typical samples of aluminum, iron with yellow coating, iron with green coating are adopted to simulate different objects on the sea. By measuring the parameters of amplitude ratio P, phase retardation θ, and completely depolarized coefficient ω_d on the platform, which stand for the surface property of the material, we can testify the accuracy of the idea. The experiments result shows that the measured P and θ values are consistent with Fresnel equations, while for ω_d , the value of seawater differs from that of the other two coating samples dramatically. As a result, it is feasible to discriminate coating target on the sea by measuring the depolarization characteristics.

Computational modeling in medical image analysis may play a critical role in surgical treatment and therapy. Many hand-crafted feature-extraction and learning based methods have been proposed for automatic brain tumor segmentation and patient survival prediction from MRI. This work first reviews few of these methods. We then discuss our recent experience with global challenge in developing computational methods for state-of-the-art brain tumor segmentation algorithms and patient survival prediction analysis.

The ability to rapidly identify symmetry and anti-symmetry is an essential attribute of intelligence. Symmetry perception is a central process in human vision and may be key to human 3D visualization. While previous work in understanding neuron symmetry perception has concentrated on the neuron as an integrator, here we show how the coincidence detecting property of the spiking neuron can be used to reveal symmetry density in spatial data. We develop a method for synchronizing symmetry-identifying spiking artificial neural networks to enable layering and feedback in the network. We show a method for building a network capable of identifying symmetry density between sets of data and present a digital logic implementation demonstrating an 8x8 leaky-integrate-and-fire (LIF) symmetry detector in a field programmable gate array. Our results show that the efficiencies of spiking neural networks can be harnessed to rapidly identify symmetry in spatial data with applications in image processing, 3D computer vision, and robotics.In conclusion, we have presented a novel algorithm for finding a scalar field representing the symmetry of points in a multi-dimensional space. We have shown how time synchronization in the input values of spiking neural networks, with the appropriate choice of threshold and spike period, results in the identification of output neurons along points of high symmetry density to the network inputs. We have demonstrated an implementation of the symmetry selective LIF neural network in common hardware with a high speed, 2.8 MHz identification of symmetry points in an 8x8 Manhattan metric space. Our results show that utilizing only the delay and coincidence detecting properties of a single layer of neurons in spiking neural networks naturally lead to effective symmetry identification. A greater understanding of symmetry perception in artificial intelligences will lead to systems with more effective pattern visualization, compression, and goal setting processes.

New control techniques are required to utilize the full potential of next generation high-energy high-repetition-rate pulses lasers while ensuring their safe operation. During automated optimization of an experiment, the control system is required to identify and reject unsafe laser configurations proposed by the optimizer. Using conventional physics codes render impossible when applied to a high energy laser system with 1ms or less time between shots, and also including laser fluctuations and drift. To mitigate this, we are using a deep Bayesian neural network to map the laser’s input power spectrum to its output power spectrum and demonstrate the speed of this approach. The Bayesian neural network can provide an estimate of its own uncertainty as a function of wavelength. A recently developed algorithm enables the uncertainty to be calculated inexpensively using multiple dropout layers inserted into the model. The uncertainty estimates are used by an active learning algorithm to improve the accuracy of the model and intelligently explore the input domain.

Visual object tracking plays a significant role in our daily life such as intelligent transportation and surveillance. However, an accurate and robust object tracker is hard to be obtained as target objects often go through huge appearance changes caused by deformation, abrupt motion, background clutter and occlusion. In this paper, we combine features extracted from deep convolutional neural networks pretrained on object recognition datasets with color name features and histogram of oriented gradient features skillfully to improve tracking accuracy and robustness. The outputs of the convolutional layers encode the senior semantic information of targets and such representations are robust to great appearance variations while their spatial resolution is too coarse to precisely locate targets. In contrast, color name features connected at the back of HOG features could provide more precise localization but are less invariant to appearance changes. We first infer the response of the convolutional features and HOG-CN features respectively, then make a linear combination of them. The maximum value of the result could represent the accurate localization of the target. We not only compare the tracking results of adopting a single feature alone, showing that the performance of them is inferior to ours, but also analyze the effect of exploiting features extracted from different convolutional layers on the tracking performance. What’s more, we introduce the adaptive target response map in our tracking algorithm to keep the target from drifting as much as possible. Extensive experimental results on a large scale benchmark dataset illustrates outstanding performance of the proposed algorithm.

Bridging the semantic gap between the low level visual features extracted by computers such as color, texture or shape and high level semantic concepts perceived by humans is the main challenge in the aim of increasing the precision of semantic results into Content-Based Image Retrieval (CBIR). This challenge has been approached with the technique known as Relevance Feedback (RF). The technique of RF can be applied through two methods, biased subspace learning or query movement. The method of query movement is based on Rocchio algorithm. In this paper, we present a new optimization to technique of Relevance Feedback through query movement to develop a CBIR system with better semantic precision. We make a modification to the input images color channels composition in the additive color space (Red, Green, Blue) and perceptual additive color space (Hue, Saturation, Value), through the images representation with human photopic vision behavior, which provides the semantic perception of the colors. With the proposed representation we obtained a more accurate behavior of the Color Histogram (CH), Color Coherence Vector (CCV) and Local Binary Patterns (LBP) descriptors in Rocchio algorithm, thus, a query movement oriented more to the semantics of the user. The optimization performance was measured with a subset of 137 classes with 100 images each one from Caltech256 object database. The results show a significant improvement in the semantic precision in comparison to the P. Mane RF method with prominent features, as well as the performance of CBIR systems without RF using the mentioned descriptors.

At the National Ignition Facility (NIF), the world’s largest most energetic laser facility with over 40,000 optics, image processing algorithms are essential to position 192 laser beams onto a mm scale fusion target. In its most general sense, an image processing algorithm consists of a series of steps which produce a desired outcome. When humans set out to solve specific tasks, they often exploit their own experience and intuition by decomposing the problem into subproblems. They then apply and develop techniques which are woven together in the end to maximize performance and minimize uncertainty. The fact that many options are available in each of these steps, can lead to a large variety of recipes from which one can choose. Consequently, the chosen solution is not unique, but multiple solutions evolve when combining the multiple choices from each step. The fact that some of the steps are nonlinear gives rise to interesting performance tradeoffs in arriving to the final solution. This paper will explore and illustrate various approaches to solving a specific image processing problem.

In areas such as computer vision, the content recognition of an image is a topic of interest in applications such as search engines, biometric security and autonomous cars, among others, since the computer must recognize all the objects that an image can have, which arises as the challenge of localizing and classifying different objects inside a single image in an efficient way. In recent years, this challenge has been approached with the use of region-based convolutional neuronal networks (R-CNN) which are systems that learn to recognize different objects by their representation in a series of images. The proposal of regions is essential for the performance of R-CNN when locating the individual objects of the image with accuracy and in the shortest time. In this article we propose a modification to a method for region proposal based on the density of SIFT like feature points that describe the objects within the image. The selection of regions is made through a decision based on the values of the cumulative distribution function of the normal distribution constructed using points density. The obtained results show a significant reduction in the processing time required for the localization of objects; having slight variations in the classification accuracy with respect to using methods such as KDRP and selective search.

Recent advances in convolution neural networks have shown promising results for the challenging task of filling large missing regions in an image with semantically plausible and context aware details. These learning-based methods are significantly more effective in capturing high-level features than prior techniques, but often create distorted structures or blurry textures inconsistent with existing areas. This is mainly due to ineffectiveness of convolutional neural networks in explicitly borrowing or copying information from distant locations. Motivated by these observations, we use a convolution neural networks architecture with Atrous Spatial Pyramid Pooling module, which can obtain multi-scale objection information, to be our inpainting network. We also use global and local Wasserstein discriminators that are jointly trained to distinguish real images from completed ones. We evaluate our approach on four datasets including faces (CelebA) and natural images (Paris Streetview, COCO, ImageNet) and achieved state-of-the-art inpainting accuracy.

In recent years, applications in the processing of images have increased, in which the objective is to improve their quality (of the image and a correct human interpretation or to facilitate the search for information in the processed image. Motivation to use, the use of new technologies to solve computational problems with high-performance computing systems that have become indispensable, when it is required to have a response in a shorter time of the result. At present, there are several systems among them, computer clusters and hybrid CPU-GPU systems that each have different characteristics and capabilities that can be applied to multiple problems. Image processing is an area that requires solving applications in a shorter time (real time), which is applicable in various areas of scientific research. The high-performance systems take advantage of all the computational processing resources using tools that measure the time of uses of the core, in which the cores are specially designed to perform matrix operations that run in parallel, this is by the number of nuclei that have the computer system. Also, you can choose the distribution of code and execution in this system many-core: the one, how, and where, the program will be executed in the many-core. [1] ADAPTEVA. The Parallella Board, User´s Guide. URL: http://www.adapteva.com/parallella-board/ [2] BERNSTEIN A. J. Analysis for programs for parallel processing. Electronic Computer, IEEE Transactions on Electronic Computers, Vol. EC-15, no. 5, pp. 757-763, Oct,1966. [3] LAMPORT L.. How to make a multiprocessor computer that correctly executes multiporcess programs. Computers, IEEE Transactions on Computers, Vol. C-28, no.9, pp. 690-691, Sept. 1979. [4] SARITA V. ADVE AND HANS-J. BOEHM. Memory Models: A Case for Rethinking Parallel Languagues and Hardware. Communications of the ACM, Vol. 53, no. 8, pp. 90–101, Aug, 2010. [5] HU WIE WU AND XIA PEISU. Out-of-order Execution in Sequentially Consistent Shared-Memory System: Theory and Experiments. Jornal of Computer. Science and Technology, Vol.13, no.2, Mar. 1998.

Motivated by recent research on asymmetric cryptosystems, a novel asymmetric scheme for image encryption that uses double random-decomposition technique in the fractional Fourier transform domain is proposed. The scheme endures the Special Attack as against conventional asymmetric cryptosystems based on phase-truncated Fourier transform (PTFT), and equal modulus decomposition. In the proposed scheme, an input image is bonded with a random phase mask and then it is subjected to a fractional Fourier transform. The resulting image is decomposed into two components using the random-decomposition technique. One of them will act as the first private key and the other component is subjected to the second fractional Fourier transform followed by another random-decomposition. Again, two new components are obtained, one will act as the second private key and the other is phase-truncated before subjecting it to LU decomposition followed by affine transform to get the encrypted image. The new scheme possesses enlarged key-space consisting of private keys obtained from random-decomposition, orders of fractional Fourier transform, affine transform parameters and permutation matrix of LU decomposition, thereby having a much greater capability to resist brute force attack. A sensitivity analysis has been carried out with respect to the encryption parameters. In addition to its resistance to the Special Attack, the scheme is immune to the basic attacks such as known-plaintext attack, chosen-plaintext attack, ciphertext-only attack, by virtue of its asymmetric nature. The above analysis along with statistical analysis through 3D plots and correlation distribution establish the strength of the proposed cryptosystem.

Random-phase-based optical image encryption techniques have drawn a lot of attention in recent years. However, in this contribution those schemes have been demonstrated to be vulnerable to chosen-plaintext attack (CPA) by employing the deep learning strategy. Specifically, by optimizing the parameters, the chosen deep neural network (DNN) can be trained to learn the sensing of an optical cryptosystem and thus get the ability to reconstruct any plaintext image from its corresponding ciphertext. A set of numerical simulation results have been further provided to shown its ability on cracking not only the classical double random phase encryption (DRPE), but also the tripe random-phase encryption (TRPE).

This paper addresses the problem of motion analysis performed from digital data captured by a network of motion sensors scattered over a 3D+T field of interest. The digital signals captured in the field are transmitted through a telecommunication network to be processed in a remote monitoring center by an artificial intelligent system based on a dual control involving both deep learning and model based algorithms. Motion analysis proceeds through consecutive steps as detection, motion-oriented classification, parameter estimation, tracking and building trajectories. Eventually, the system can detect and predict abnormalities, incidents and accidents. In all current applications, it is commonly thought that motion analysis is to be performed from data streams captured from multiple video-cameras distributed in a network following a so-called “camera-everywhere” approach. A basic observation of how animal biology proceeds shows that information analyzed by the cortex usually originates from one global eye and from a network of sensors non-uniformly distributed over the the entire skin. Telecommunications are performed through a network of nerves that acts as a bundle of telephone lines that connects each sensor located in the eye and in the skin to specific and dedicated areas of the cortex. This paper shows the relevance of this natural way to perform full motion analysis from a network of motion sensor distributed over a field of interest where the 3D+T motion analysis is performed. Using all video-cameras in a network involves two main drawbacks. First, this setting requires to transmit over a telecommunication network an amount of information of which more than ninety nine percent of the content is useless for motion analysis. Second, additional software algorithms that are consuming time and resources have to be implemented to extract the motion information from the video signal and build the trajectories. Video cameras are in fact useful for both pattern recognition and motion disambiguation, and therefore, should be be limited in number and located at key spots or on robots moving in the field. Eventually, a central station implements the motion analysis algorithm. This paper describes the application with a three layer scheme and compares both approaches namely composed of all video-cameras or all motion sensors. Comparisons are presented between two schemes, namely multiple video cameras and sensor networks, in term of networking and processing the motion information in the remote monitoring center.

RGBD cameras provide images of depth and 3D information of physical objects. These cameras, which are usually based on Time of Flight techniques, have become an alternative for detecting superficial defects in industrial production. Traditionally, the measurement of changes in thickness of a transparent flat glass has been carried out by means of point like techniques. Thus, for the measurement of an extended surface of the glass a scanning process is necessary. This drawback can be improved using cameras RGBD like the Kinect V2. In this article, a novel method for detecting the change in thickness of an area of interest of a glass is described. This proposal is based on the refraction of the light when it goes from air to glass, i.e., it is based on the change of direction and velocity of the light due to the change of refraction index between the media. This phenomenon is evident in the RGBD images of a known scene when a transparent flat glass is put before its background. The changes in the RGBD images of the scene with and without glass are caused by the increment in time of flight by the refraction in the glass. We obtained a correlation between the thicknesses of different glasses and the depth measurements for a controlled scene, by using an experimental setup with a Kinect V2 as RGBD camera. We processed the data using a K-means algorithm to determine the change of thickness in transparent flat glasses.

In computational vision has a high computational cost, although, some algorithms had been implemented to get image features, that allow assorting, object and face recognition and so on. Some solutions have been developed in computers, DSP and GPU those that are not optimal with time. In order to improve the performance of these algorithms, we are implementing the SURF algorithm in embedded systems (FPGA) and applied to non-controller environments that require real-time response. In this work we development a SURF algorithm in order to improve time processing in video and image processing, we use an FPGA to apply that algorithm, we compare the time processing with different devices and the features found it into the images, this features will be invariant to scale, rotation and lighting, the SURF algorithm localize the interest points (features), its is using in facial recognition, object detection, stereo vision and so on. This algorithm has a high computational cost because of use a lot of data, in order to reduce the high cost we implemented LUTs and reduce time with code. With this work we try to find the best way to implement the algorithm into embedded systems, in order to use in non-controller environments and robots autonomous.

Human action recognition in indoor environment can prove to be very crucial in avoiding serious accidents and (or) damage. Application domain spans from monitoring the actions of solitary elders or persons with disabilities to monitoring persons working alone in a chamber or in isolated industry environment. These scenarios demand an automatic near real-time activity recognition and alert to save life and assets. In this work, considering the fact that the sensing modality should be capable of working round the clock in a non-intrusive manner, we have opted for thermal infrared camera, which captures the heat emitted by objects in the scene and generates an image. Motivated by the recent success of convolutional neural networks (CNN) for human action recognition in IR images, we extend this work by incorporating one additional dimension i.e. the temporal information. In this work, we have designed and implemented a 3D-CNN for learning the spatial as well as the sequential features in the thermal IR videos. In this work, eight action classes are considered - Walking, Standing, Falling, Lying, Sitting, Falling from chair, Sitting up (recovering from fall from sitting posture), Getting up (recovering from fall from lying posture). To evaluate the proposed framework, infrared (IR) videos of different actions were generated in three diverse environments of home – inside study room, inside a bedroom and in the garden. The dataset comprised of 2641 and 894 IR videos for training and testing respectively, each of half a second duration performed by more than 50 volunteers. We have designed and implemented 3D-CNN, comprising of two blocks, each of two convolution and one max pool layer, which automatically constructs features from raw data incorporating both spatial and temporal information to learn actions. Network parameters are learned using back-propagation algorithm and the learning is supervised. Experimental results indicate 85% classification accuracy on 894 complex test videos of the proposed Spatio-Temporal Deep Learning architecture on the IR action dataset.

We have included a Parallel Aligned Liquid Crystal on Silicon (PA-LCoS) microdisplay in a Holographic Data Storage System (HDSS). This novel display, widely accepted as Spatial Light Modulator (SLM), presents some advantages and disadvantages. One of these disadvantages is the anamorphic and frequency dependent effect. In this work we want to test this effect and see its effects in the complete optical process involved in the HDSS. We will use stripe-based patterns with different orientation (vertical and horizontal). To check the limits, we will increase the data density by decreasing the minimum stripe width. For evaluating the degradation suffered by the data page, we use the Bit Error Rate (BER) as figure of merit. We make a BER calculation from the statistical analysis of the histogram. In addition to the anamorphic effects we evaluate the degradation effects introduced by the non-uniformity in the illumination. To this goal we divide the image in several regions that are processed in the same way that the entire image. The error analysis of the entire optical system is useful for its calibration and fine adjustment. Once we have characterized the experimental setup we introduce the holographic material. Thus, by making the same analysis, we can evaluate the errors introduced by the material. As holographic material we use Polyvinyl Alcohol Acrylamide (PVA/AA), that has been characterized and developed in previous works by our group.

We derive the graph-analytic representation of influence of the higher-order nonlinearities (including the third and fourth ones) of holographic recording on the associative properties of second-order holograms defined as ‘holograms between two holograms’. It is shown that the higher-order nonlinearities of the amplitude response of a hologram do not only cause the noise contribution into conjugate associative response, but in some scenarios of formation of the second-order hologram just predetermine reconstruction of this response. Using the proposed technique, we analyze various cases of imposed (sequential) record of partial signals within the framework of the model of multiple diffraction at hologram structure. We discuss the cases when reconstruction of the conjugate associative response is provided by the presence of the hierarchy of combination pseudogratings, rather than by the ‘direct’ interference mechanism that lies in the base of conventional holographic associative storage, using ghost-image holograms. The represented results expand considerably functional feasibilities of the phase-conjugation associative memories on the base of static holograms.

We represent the double phase conjugation technique for uncorrelated complex optical signals at arbitrarily different (incommensurable) wave lengths for implementation of long-term (archive) storage, i.e. coupling and mutual associative reconstruction of such signals. The essence of the proposed approach consists in exploiting natural recording nonlinearity of a static hologram that results in formation of the combined (summation) pseudogratings corresponding to the quadratic component of the amplitude response of a static nonlinearly recorded hologram. In contrast to earlier (real-time holography) version of the double phase conjugation using photorefractive crystals, we use for the each signal wave collimated (plane) reference wave, so that the wave vectors of two reference waves are strictly opposite to each other. It is shown that under these conditions nonlinear mixing of two sets of cross-gratings results in formation of the complete set of pseudogratings constituting a second-order hologram defined, following H.J. Caulfield, as ‘a hologram between two (linear) holograms’. Being read out by any of two stored signals or its incomplete/distorted version (in absence of the reference waves, just as in a photorefractive prototype), a hologram reconstructs the phase-conjugate replica of the second signal (heteroassociative response) at the wavelength of the readout beam, with predictable on the wavelength ratio scaling and angular shift from the nominal position. Especial attention is paid to determination of the experimental conditions for providing the combination pseudogratings to be thin (by applying the Klein’s parameter), if even the partial cross-gratings are thick (volume), by proper choice of the angular conditions of the experiment. If this condition is violated, the Bragg selectivity can hinder heteroassociative reconstruction.

We represent the straightforward techniques for differentiation of the phase structures of light beams of various origin but having similar intensity envelopes using doughnut-like beams as an example. Namely, we compare the phase structure of the set of beams: (i) a ring beam with smooth (vortexless) wavefront but with artificially introduced central amplitude zero, with Gaussian radial intensity distribution; (ii) Laguerre-Gaussian mode LG1₀⁺¹ with the central vortex; (iii) combined beam assembled from uncorrelated weighed Laguerre-Gaussian modes LG¹₀ and LG¹₁ with the central screw dislocation and with the ring edge dislocation of the spatial coherence function; (iv) combined beam assembled from uncorrelated Hermite-Gaussian modes HG¹₀ and HG₁⁰ ; (v) combined beam assembled from correlated but orthogonal in polarization Hermite-Gaussian modes HG¹₀ and HG₁⁰ ; (vi) combined beam assembled from uncorrelated and orthogonal in polarization Hermite-Gaussian modes HG¹₀ and HG₁⁰ . Experimental analysis and comparison of the phase structures (i) and (ii) can be performed using a common interference technique with off-axis reference wave. Other mentioned cases cannot be analyzed by applying this technique. To differentiate the corresponding phase structures and associated singularities, we attract the united technique based on edge diffraction and use of an opaque strip screen placed at the analyzed beam. In cases (v) and (vi), this technique is added by 2D Stokes polarimetry. The proposed techniques provide reliable diagnostics of common optical vortices, vortices of the spatial correlation functions, polarization singularities of completely (but inhomegeneously) polarized light beams, and the singularities of the complex degree of polarization from typical bending or a half-period shift of the Young’s interference fringes at the shadow of the strip screen.

An overview of existing technical solutions for creating visualization systems for light microscopes is presented. Theoretical and practical aspects of the design of optical systems are considered taking into account the projection of the image of the object on the electronic receiver. The analysis of the basic tendencies in development of modern visualization systems for light microscopes is made. It is necessary to fulfill the basic requirements that ensure correlation when observing images in the eyepieces and using a visualization system. System must to transfer to the receiver linear image of the microscope corresponding to the resolution of the used objective.

Systems of remote control with unidirectional communication channel, made using KeeLoq technology, have found wide application in such modern devices as car security systems, identification systems, systems for restricting access to premises, electronic locks, process control, etc. The paper describes a comparative analysis of KeeLoq technologies [1] and SmtLoq. The SmtLoq algorithm uses a set of random numbers in each transmission, and does not use serial numbers, factory and secret keys, counters. In SmtLoq technology, information about the transferred command is determined from the correlation analysis of the transmitted sequence of random numbers. The peculiarity of the new SmtLoq technology is protection from interception in a row even of several successive control commands. The minimum size of the control packet in the SmtLoq algorithm is four bytes of random numbers and one byte of the checksum. The maximum probability of guessing the next command decreases with the increase in the number of random data in the control command packet. The probability of repetition of a set of random data in the transmitted packet is extremely small, thus, the single use of a certain set of random data in a packet ensures the uselessness of known computer methods of code selection. The lack of the use of keys and serial numbers provides an opportunity for each user to create a unique codec-decoder pair from arbitrary eight-bit chips with appropriate firmware for the encoder and decoder.

The paper considers results of design and simulation of continuously logical cells (CLC) based on current mirrors (CM) with functions of preliminary analogue processing for image intensity transformation and coding for construction of mixed image processors (IP) and neural networks (NN). For such IP and NN with vector or matrix parallel inputsoutputs, it is needed active basic photosensitive cells with an extended electronic circuit, which are considered in paper. Such CLC has a number of advantages: high speed and reliability, simplicity, small power consumption, high integration level for linear and matrix structures. We show design of CLC variants for photocurrents transformation and coding and their various possible implementations and simulations. The basic element of such cells is a scheme that implements the operation of a bounded difference of continuous logic. Using a set of such circuits implemented on traditional CMOS technology, we consider generalized decomposition and other methods for designing cells for nonlinear conversion of the photocurrent intensity, which makes it easy to realize the required nonlinear conversion function. Selection of the appropriate parameters, which can be specified as constructive constants or as parameters for external control, allows changing type of synthesized functions. We also consider the applications of such parallel matrix arrays for the creation of advanced IP and NN. We show the need for various types of converting and coding the photocurrents intensity in such parallel systems and sensory devices, especially for the implementation of various types of activation functions in the hardware implementations of neural networks. Such cells consist of several dozen CMOS transistors, have low power supply voltage (1.8 ÷ 3.3V), the range of an input photocurrent is 0.1÷24μA, the transformation time is less than 1 μs, low power consumption (microwatts). We also consider the cells for ADC after the intensity conversion. Each channel consists of several digital-analog cells (DC). The amount of DC is not exclusive to the number of digits of the formed code, and for an iteration type, only one cell of DC, complemented by the device of selection and holding (SHD), is required. One channel of ADC with iteration is based on one DC and SHD, and it has only 35 CMOS transistors. The general power consumption of the ADC with iteration is only 50÷100μW, if the maximum input current is 4μA. The ADCs with 6-8 bit binary or Gray codes have good dynamic characteristics (frequency of digitization even for 1.5μm CMOS-technologies is 10MHz and more). In such ADCs, easily parallel code can be realized. The circuits and the simulation results of their design with OrCAD are shown. The CLC and ADC on current mirrors open new prospects for realization of linear and matrix IP and NN with MIMO-operands.

In this paper will be presented new opportunities for developing innovative solutions based on blockchain technologies in creation of new data management algorithms, especially those focused on the distribution and processing of data in layered structures, where independent instances can fully control and verify all operations. Later an attempt will be made to generalize protocols, which will be proposed for management activities, towards their extension and the possibility of using them in hierarchical structures. Blockchain technology should enable the development of decentralized management protocols that will allow complete verifiability of all operations by all authorized parties that can collect, process and transmit such information. In practice such protocols will give the opportunity to create an effective and secure data sharing protocols in structures, where all authorized entities are equal and can independently verify the type of operations and instances that process such data.

In this paper will be presented new possible applications of cognitive vision or information systems for secure information distribution and management in Cloud infrastructures. Cognitive systems are dedicated to visual data interpretation and semantic analysis. Among new possibilities of application arise secure data distribution and management in distributed computer infrastructures like Cloud and Fog computing. In this paper cryptographic approaches for data division will be described, which allow to improve processes of secure and efficient information fusion with application of cognitive procedures. A few possible application of proposed approaches for visual information encryption and distribution will be also described.

Automatic vehicle type classification plays a significant role in security, traffic control and autonomous driving applications. Thermal infrared (IR) cameras operating even in complete darkness and adverse weather conditions emerge as a potential sensing modality for such challenging outdoor applications. However, automated vehicle type classification in infrared imagery still poses significant challenges due to high variability of vehicle signature in infrared band leading to high intra-class variation and low inter-class variation. To address these issues, we demonstrate the use of local features represented in a bag of words framework. In this work, we present comparative analysis of two feature detectors, MSER – a sparse region based detector and uniform dense sampling of points in the image across multiple scales (termed dense). A bag of features (BoF) framework based on SURF feature descriptor and SVM classifier for vehicle type classification are evaluated on a thermal infrared (TIR) vehicle dataset. A number of variations are present in the TIR vehicle dataset - scale variation, pose variation and partial visibility of vehicles captured under varied environmental conditions. The dataset contains four vehicle categories commonly plying on Indian roads, Bike, Autorickshaw, Car and Heavy vehicle. The performance of the designed vehicle type classification framework was evaluated using performance metrics, classification accuracy and confusion matrix. The optimized sparse MSER and dense BoF framework demonstrated decent classification accuracies of 85.7% and 93% respectively for automatic vehicle type classification on the thermal infrared vehicle dataset.