This PDF file contains the front matter associated with SPIE Proceedings Volume 9477 including the Title Page, Copyright information, Table of Contents, Introduction, and Conference Committee listing.

We originally proposed and experimentally demonstrated the targeted-light delivery capability of so-called Wave-guided Optical Waveguides (WOWs) three years ago. As these WOWs are maneuvered in 3D space, it is important to maintain efficient light coupling through their integrated waveguide structures. In this work we demonstrate the use of real-time diffractive techniques to create focal spots that can dynamically track and couple to the WOWs during operation in a volume. This is done by using a phase-only spatial light modulator to encode the needed diffractive phase patterns to generate a plurality of dynamic coupling spots. In addition, we include our proprietary GPC Light Shaper before the diffractive setup to efficiently illuminate the rectangular shaped spatial light modulator by a Gaussian laser beam. The method is initially tested for a single WOW and we have experimentally demonstrated dynamic tracking and coupling for both lateral and axial displacements of the WOWs. The ability to switch from on-demand to continuous addressing with efficient illumination leverages our WOWs for potential applications in near-field stimulation and nonlinear optics at small scales.

Face recognition is a critical tool used in almost all major biometrics based security systems. But recognition, authentication and liveness detection of the face of an actual user is a major challenge because an imposter or a non-live face of the actual user can be used to spoof the security system. In this research, a robust technique is proposed which detects liveness of faces in order to counter spoof attacks. The proposed technique uses a three-dimensional (3D) fast Fourier transform to compare spectral energies of a live face and a fake face in a mathematically selective manner. The mathematical model involves evaluation of energies of selective high frequency bands of average power spectra of both live and non-live faces. It also carries out proper recognition and authentication of the face of the actual user using the fringe-adjusted joint transform correlation technique, which has been found to yield the highest correlation output for a match. Experimental tests show that the proposed technique yields excellent results for identifying live faces.

A spatial domain optimal trade-off Maximum Average Correlation Height (OT-MACH) filter has been previously developed and shown to have advantages over frequency domain implementations in that it can be made locally adaptive to spatial variations in the input image background clutter and normalised for local intensity changes. In this paper we compare the performance of the spatial domain (SPOT-MACH) filter to the widely applied data driven technique known as the Scale Invariant Feature Transform (SIFT). The SPOT-MACH filter is shown to provide more robust recognition performance than the SIFT technique for demanding images such as scenes in which there are large illumination gradients. The SIFT method depends on reliable local edge-based feature detection over large regions of the image plane which is compromised in some of the demanding images we examined for this work. The disadvantage of the SPOTMACH filter is its numerically intensive nature since it is template based and is implemented in the spatial domain.

The fringe-adjusted joint transform correlation (FJTC) technique has been widely used for real-time optical pattern recognition applications. However, the classical FJTC technique suffers from target distortions due to noise, scale, rotation and illumination variations of the targets in input scenes. Several improvements of the FJTC have been proposed in the literature to accommodate these problems. Some popular techniques such as synthetic discriminant function (SDF) based FJTC was designed to alleviate the problems of scale and rotation variations of the target, whereas wavelet based FJTC has been found to yield better performance for noisy targets in the input scenes. While these techniques integrated with specific features to improve performance of the FJTC, a unified and synergistic approach to equip the FJTC with robust features is yet to be done. Thus, in this paper, a robust FJTC technique based on sequential filtering approach is proposed. The proposed method is developed in such a way that it is insensitive to rotation, scale, noise and illumination variations of the targets. Specifically, local phase (LP) features from monogenic signal is utilized to reduce the effect of background illumination thereby achieving illumination invariance. The SDF is implemented to achieve rotation and scale invariance, whereas the logarithmic fringe-adjusted filter (LFAF) is employed to reduce the noise effect. The proposed technique can be used as a real-time region-of-interest detector in wide-area surveillance for automatic object detection. The feasibility of the proposed technique has been tested on aerial imagery and has observed promising performance in detection accuracy.

Human detection has gained considerable importance in aggravated security scenarios over recent times. An effective security application relies strongly on detailed information regarding the scene under consideration. A larger accumulation of humans than the number of personal authorized to visit a security controlled area must be effectively detected, amicably alarmed and immediately monitored. A framework involving a novel combination of some existing techniques allows an immediate detection of an undesirable crowd in a region under observation. Frame differencing provides a clear visibility of moving objects while highlighting those objects in each frame acquired by a real time camera. Training of a correlation pattern recognition based filter on desired shapes such as elliptical representations of human faces (variants of an Omega Shape) yields correct detections. The inherent ability of correlation pattern recognition filters caters for angular rotations in the target object and renders decision regarding the existence of the number of persons exceeding an allowed figure in the monitored area.

The presented paper is of PC modeling of scaled object recognition with the help of invariant correlation filters for optical correlators. The object database consists of objects of true and false classes with different changes of scale. The recognition process consits of two stages-multiclass recognition and geometrical change recognition with the help of filters of different types. The results of modeling present data on comparison of different combination of filter types.

The nonlinearity inherent in four-wave mixing in photorefractive (PR) materials is used for adaptive filtering. Examples include script enhancement on a periodic pattern, scratch and defect cluster enhancement, periodic pattern dislocation enhancement, etc. through intensity filtering image manipulation. Organic PR materials have large space-bandwidth product, which makes them useful in adaptive filtering techniques in quality control systems. For instance, in the case of edge enhancement, phase conjugation via four-wave mixing suppresses the low spatial frequencies of the Fourier spectrum of an aperiodic image and consequently leads to image edge enhancement. In this work, we model, numerically verify, and simulate the performance of a four wave mixing setup used for edge, defect and pattern detection in periodic amplitude and phase structures. The results show that this technique successfully detects the slightest defects clearly even with no enhancement. This technique should facilitate improvements in applications such as image display sharpness utilizing edge enhancement, production line defect inspection of fabrics, textiles, e-beam lithography masks, surface inspection, and materials characterization.

Thermal images are exploited in many areas of pattern recognition applications. Infrared thermal image segmentation can be used for object detection by extracting regions of abnormal temperatures. However, the lack of texture and color information, low signal-to-noise ratio, and blurring effect of thermal images make segmenting infrared heat patterns a challenging task. Furthermore, many segmentation methods that are used in visible imagery may not be suitable for segmenting thermal imagery mainly due to their dissimilar intensity distributions. Thus, a new method is proposed to improve the performance of image segmentation in thermal imagery. The proposed scheme efficiently utilizes nonlinear intensity enhancement technique and Unsupervised Active Contour Models (UACM). The nonlinear intensity enhancement improves visual quality by combining dynamic range compression and contrast enhancement, while the UACM incorporates active contour evolutional function and neural networks. The algorithm is tested on segmenting different objects in thermal images and it is observed that the nonlinear enhancement has significantly improved the segmentation performance.

In this paper we present a technique to detect pedestrian. Histogram of gradients (HOG) and Haar wavelets with the aid of support vector machines (SVM) and AdaBoost classifiers show good identification performance on different objects classification including pedestrians. We propose a new shape descriptor derived from the intra-relationship between gradient orientations in a way similar to the HOG. The proposed descriptor is a two 2-D grid of orientation similarities measured at different offsets. The gradient magnitudes and phases derived from a sliding window with different scales and sizes are used to construct two 2-D symmetric grids. The first grid measures the co-occurence of the phases while the other one measures the corresponding percentage of gradient magnitudes for the measured orientation similarity. Since the resultant matrices will be symmetric, the feature vector is formed by concatenating the upper diagonal grid coefficients collected in a raster way. Classification is done using SVM classifier with radial basis kernel. Experimental results show improved performance compared to the current state-of-art techniques.

The human brain has the capability to process high quantities of data quickly for detection and recognition tasks. These tasks are made simpler by the understanding of data, which intentionally removes redundancies found in higher dimensional data and maps the data onto a lower dimensional space. The brain then encodes manifolds created in these spaces, which reveal a specific state of the system. We propose to use a recurrent neural network, the nonlinear line attractor (NLA) network, for the encoding of these manifolds as specific states, which will draw untrained data towards one of the specific states that the NLA network has encoded. We propose a Gaussian-weighted modular architecture for reducing the computational complexity of the conventional NLA network. The proposed architecture uses a neighborhood approach for establishing the interconnectivity of neurons to obtain the manifolds. The modified NLA network has been implemented and tested on the Electro-Optic Synthetic Vehicle Model Database created by the Air Force Research Laboratory (AFRL), which contains a vast array of high resolution imagery with several different lighting conditions and camera views. It is observed that the NLA network has the capability for representing high dimensional data for the recognition of the objects of interest through its new learning strategy. A nonlinear dimensionality reduction scheme based on singular value decomposition has found to be very effective in providing a low dimensional representation of the dataset. Application of the reduced dimensional space on the modified NLA algorithm would provide fast and more accurate recognition performance for real time applications.

One of the most difficult challenges of working with LiDAR data is the large amount of data points that are produced. Analysing these large data sets is an extremely time consuming process. For this reason, automatic perception of LiDAR scenes is a growing area of research. Currently, most LiDAR feature extraction relies on geometrical features specific to the point cloud of interest. These geometrical features are scene-specific, and often rely on the scale and orientation of the object for classification. This paper proposes a robust method for reduced dimensionality feature extraction of 3D objects using a volume component analysis (VCA) approach.1 This VCA approach is based on principal component analysis (PCA). PCA is a method of reduced feature extraction that computes a covariance matrix from the original input vector. The eigenvectors corresponding to the largest eigenvalues of the covariance matrix are used to describe an image. Block-based PCA is an adapted method for feature extraction in facial images because PCA, when performed in local areas of the image, can extract more significant features than can be extracted when the entire image is considered. The image space is split into several of these blocks, and PCA is computed individually for each block. This VCA proposes that a LiDAR point cloud can be represented as a series of voxels whose values correspond to the point density within that relative location. From this voxelized space, block-based PCA is used to analyze sections of the space where the sections, when combined, will represent features of the entire 3-D object. These features are then used as the input to a support vector machine which is trained to identify four classes of objects, vegetation, vehicles, buildings and barriers with an overall accuracy of 93.8%

For rigid objects and fixed scenes, current machine vision technology is capable of identifying imagery rapidly and with specificity over a modest range of camera viewpoints and scene illumination. We applied that capability to the problem of runway identification using video of sixteen runway approaches at nine locations, subject to two simplifying assumptions. First, by using approach video from just one of the several possible seasonal variations (no snow cover and full foliage), we artificially removed one source of scene variation in this study. Secondly, by not using approach video at dawn and dusk, we limited the study to two illumination variants (day and night). We did allow scene variation due to atmospheric turbidity by using approach video from rainy and foggy days in some daytime approaches. With suitable ensemble statistics to account for temporal continuity in video, we observed high location specificity (<90% Bayesian posterior probability). We also tested repeatability, i.e., identification of a given runway across multiple videos, and observed robust repeatability only if illumination (day vs. night) was the same and approach visibility was good. Both specificity and repeatability degraded in poor weather conditions. The results of this simplified study show that geolocation via real-time comparison of cockpit image sensor video to a database of runway approach imagery is feasible, as long as the database contains imagery from about the same time of day (complete daylight and nighttime, excluding dawn and dusk) and the weather is clear at the time of the flight.

In this paper, we propose and validate a new system used to explore road assets. In this work we are interested on the vertical road signs. To do this, we are based on the combination of road signs detection, recognition and identification using data provides by sensors. The proposed approach consists on using panoramic views provided by the innovative device, VIAPIX®¹, developed by our company ACTRIS². We are based also on the optimized correlation technique for road signs recognition and identification on pictures. Obtained results shows the interest on using panoramic views compared to results obtained using images provided using only one camera.

Several pattern-matching techniques have focused on affine invariant pattern matching, mainly because rotation, scale, translation, and shear are common image transformations. In some situations, other transformations may be modeled as a small deformation on top of an affine transformation. This work presents an algorithm which aims at improving existing Fourier Transform (FT)-based pattern matching techniques in such a situation. The pattern is first decomposed into non-overlapping concentric circular rings, which are centered in middle of the pattern. Then, the FT of each ring is computed. Essentially, adding the individual complex-valued FTs provides the overall FT of the pattern. Past techniques used the overall FT to identify the parameters of the affine transformation between two patterns. In this work, it is assumed that the rings may be rotated with respect to each other, thus, parameters of transformations beyond the affine ones can be computed. The proposed method determines this variable angle of rotation starting from the FT of the outermost ring and moving inwards to the FT of the innermost ring. The variable angle of rotation provides information about the directional properties of a pattern. Two methods are investigated, namely a dynamic programming algorithm and a greedy algorithm, in order to determine the variable angle of rotation. The intuition behind this approach is that since the rings are not necessarily aligned in the same manner for different patterns, their ring FTs may also be rotated with respect to each other. Simulations demonstrate the effectiveness of the proposed technique.

The topic of constructing data-dependent dictionaries, referred to as dictionary learning, has received considerable interest in the past decade. In this work, we compare the ability of two dictionary learning algorithms, K-SVD and geometric multi-resolution analysis (GMRA), to perform image reconstruction using a fixed number of coefficients. K-SVD is an algorithm originating from the compressive sensing community and relies on optimization techniques. GMRA is a multi-scale technique that is based on manifold approximation of highdimensional point clouds of data. The empirical results of this work using a synthetic dataset of images of vehicles with diversity in viewpoint and lighting show that the K-SVD algorithm exhibits better generalization reconstruction performance with respect to test images containing lighting diversity that were not present in the construction of the dictionary, while GMRA exhibits superior reconstruction on the training data.

Pose estimation is playing the vital role in the final approach phase of two spacecraft, one is the target spacecraft and the other one is the observation spacecraft. Traditional techniques are usually based on feature tracking, which will not work when sufficient features are unavailable. To deal with this problem, we present a stereo camera-based pose estimation method without feature tracking. First, stereo vision is used to reconstruct 2.5D of the target spacecraft and a 3D reconstruction is presented by merged all the point cloud of each viewpoint. Then a target-coordinate system is built using the reconstruction results. Finally, point cloud registration algorithm is used to solve the current pose between the observation spacecraft and the target spacecraft. Experimental results show that both the position errors and the attitude errors satisfy the requirements of pose estimation. The method provides a solution for pose estimation without knowing the information of the targets and this algorithm is with wider application range compared with the other algorithms based on feature tracking.

In space rendezvous and docking missions, pose estimation of non-cooperative targets is challenging as priori information about motion and structure are unknown. The extraction and recognition are far more difficult conducted on a whole target. To solve this problem, a pose estimation method based on docking surface is proposed. The docking surfaces have more sophisticated structure with similar appearance among different countries than other surfaces. So docking surface is easy to automatically recognize or manually mark in images. In this paper, a control point representing mark information is chosen to assist with docking surface detection. The vertices of docking surface can be used to estimate pose. Firstly, binocular images are obtained by 3-D simulation technology. Then, a novel framework is proposed to detect edges of the docking surface in each image. Specifically, we detect lines in an image and group them according to the slopes. The control point is utilized to pick out the edges from the lines detected. Finally, the pose of the target is calculated by the four vertices of the docking surface. Simulation result shows that the position errors and attitude errors meet the requirement of pose estimation. The method provides a new solution for pose estimation of the non-cooperative target, and has potential significance for space rendezvous and docking technology.

In this study, we propose a numerical implementation (using a GPU) of an optimized multiple image compression and encryption technique. We first introduce the double optimization procedure for spectrally multiplexing multiple images. This technique is adapted, for a numerical implementation, from a recently proposed optical setup implementing the Fourier transform (FT)¹. The new analysis technique is a combination of a spectral fusion based on the properties of FT, a specific spectral filtering, and a quantization of the remaining encoded frequencies using an optimal number of bits. The spectral plane (containing the information to send and/or to store) is decomposed in several independent areas which are assigned according a specific way. In addition, each spectrum is shifted in order to minimize their overlap. The dual purpose of these operations is to optimize the spectral plane allowing us to keep the low- and high-frequency information (compression) and to introduce an additional noise for reconstructing the images (encryption). Our results show that not only can the control of the spectral plane enhance the number of spectra to be merged, but also that a compromise between the compression rate and the quality of the reconstructed images can be tuned. Spectrally multiplexing multiple images defines a first level of encryption. A second level of encryption based on a real key image is used to reinforce encryption. Additionally, we are concerned with optimizing the compression rate by adapting the size of the spectral block to each target image and decreasing the number of bits required to encode each block. This size adaptation is realized by means of the root-mean-square (RMS) time-frequency criterion². We have found that this size adaptation provides a good trade-off between bandwidth of spectral plane and number of reconstructed output images³. Secondly, the encryption rate is improved by using a real biometric key and randomly changing the rotation angle of each block before spectral fusion. A numerical implementation of this method using two numerical devices (CPU and GPU) is presented⁴.

Registering the 2D images is one of the important pre-processing steps in many computer vision applications like 3D reconstruction, building panoramic images. Contemporary registration algorithm like SIFT (Scale Invariant Feature transform) was not quite success in registering the images under symmetric conditions and under poor illuminations using DoF (Difference of Gaussian) features. In this paper, we introduced a novel approach for registering the images under symmetric conditions.

Thresholding techniques that account for noise are essential for the efficiency and accuracy of an optical communication or optical data storage system. Various types of noise in the system can result in error. To recover the data from the noisy signal, the error must be corrected by a fast and accurate signal processing algorithm. By considering the crosstalk effect of the neighboring channels, we have devised a multi-level thresholding method to set the threshold values based on the neighboring channel values. We compare the binary characterization performance of a neural network and the local multi-level adaptive thresholding method for decoding noisy transmission images. We show that the multi-thresholding implementation results in an average of 57.42% less binary characterization errors than the artificial neural network across twenty unique mixed noise optical conditions.

We have developed a Holographic Content Addressable Storage (HCAS) architecture. The HCAS systems consists of a DMD (Digital Micromirror Array) as the input Spatial Light Modulator (SLM), a CMOS (Complementary Metal-oxide Semiconductor) sensor as the output photodetector and a photorefractive crystal as the recording media. The HCAS system is capable of performing optical correlation of an input image/feature against massive reference data set stored in the holographic memory. Detailed system analysis will be reported in this paper.