Adaptive feature-specific imaging: a face recognition example

Compressive phase object classification using single-pixel digital holography

A single-pixel camera (SPC) is a computational imaging system that obtains compressed signals of a target scene using a single-pixel detector. The compressed signals can be directly used for image classification, thereby bypassing image reconstruction, which is computationally intensive and requires a high measurement rate. Here, we extend this direct inference to phase object classification using single-pixel digital holography (SPDH). Our method obtains compressed measurements of target complex amplitudes using SPDH and trains a classifier using those measurements for phase object classification. Furthermore, we present a joint optimization of the sampling patterns used in SPDH and a classifier to improve classification accuracy. The proposed method successfully classified phase object images of handwritten digits from the MNIST database, which is challenging for SPCs that can only capture intensity images.

Multilinear Compressive Learning

Compressive learning (CL) is an emerging topic that combines signal acquisition via compressive sensing (CS) and machine learning to perform inference tasks directly on a small number of measurements. Many data modalities naturally have a multidimensional or tensorial format, with each dimension or tensor mode representing different features such as the spatial and temporal information in video sequences or the spatial and spectral information in hyperspectral images. However, in existing CL frameworks, the CS component utilizes either random or learned linear projection on the vectorized signal to perform signal acquisition, thus discarding the multidimensional structure of the signals. In this article, we propose multilinear CL (MCL), a framework that takes into account the tensorial nature of multidimensional signals in the acquisition step and builds the subsequent inference model on the structurally sensed measurements. Our theoretical complexity analysis shows that the proposed framework is more efficient compared to its vector-based counterpart in both memory and computation requirement. With extensive experiments, we also empirically show that our MCL framework outperforms the vector-based framework in object classification and face recognition tasks, and scales favorably when the dimensionalities of the original signals increase, making it highly efficient for high-dimensional multidimensional signals.

A de-illumination scheme for face recognition based on fast decomposition and detail feature fusion

Almost all the face recognition algorithms are unsatisfied due to illumination variation. Feature with high frequency represents the face intrinsic structure according to the common assumption that illumination varies slowly and the face intrinsic feature varies rapidly. In this paper, we will propose an adaptive scheme based on FBEEMD and detail feature fusion. FBEEMD is a fast version of BEEMD without time-consuming surface interpolation and iteration computation. It can decompose an image into sub-images with high frequency matching detail feature and sub-images with low frequency corresponding to contour feature. However, it is difficult to determine by quantitative analysis that which sub-images with high frequency can be used for reconstructing an illumination-invariant face. Thus, two measurements are proposed to calculate weights for quantifying the detail feature. With this fusion technique, one can reconstruct a more illumination-neutral facial image to improve face recognition rate. Verification experiments using classical recognition algorithms are tested with Yale B, PIE and FERET databases. The encouraging results show that the proposed scheme is very effective when dealing with face images under variable lighting condition.

A coded aperture compressive imaging array and its visual detection and tracking algorithms for surveillance systems

In this paper, we propose an application of a compressive imaging system to the problem of wide-area video surveillance systems. A parallel coded aperture compressive imaging system is proposed to reduce the needed high resolution coded mask requirements and facilitate the storage of the projection matrix. Random Gaussian, Toeplitz and binary phase coded masks are utilized to obtain the compressive sensing images. The corresponding motion targets detection and tracking algorithms directly using the compressive sampling images are developed. A mixture of Gaussian distribution is applied in the compressive image space to model the background image and for foreground detection. For each motion target in the compressive sampling domain, a compressive feature dictionary spanned by target templates and noises templates is sparsely represented. An l(1) optimization algorithm is used to solve the sparse coefficient of templates. Experimental results demonstrate that low dimensional compressed imaging representation is sufficient to determine spatial motion targets. Compared with the random Gaussian and Toeplitz phase mask, motion detection algorithms using a random binary phase mask can yield better detection results. However using random Gaussian and Toeplitz phase mask can achieve high resolution reconstructed image. Our tracking algorithm can achieve a real time speed that is up to 10 times faster than that of the l(1) tracker without any optimization.

Adaptive feature-specific imaging for recognition of non-Gaussian classes

We present an adaptive feature-specific imaging (AFSI) system for application to an M-class recognition task. The proposed system uses nearest-neighbor-based density estimation to compute the (non-Gaussian) class-conditional densities. We refine the density estimates based on the training data and the knowledge from previous measurements at each step. The projection basis for the AFSI system is also adapted based on the previous measurements at each step. The decision-making process is based on sequential hypothesis testing. We quantify the number of measurements required to achieve a specified probability of error (P(e)) and we compare the AFSI system with an adaptive-conventional (ACONV) system. The AFSI system exhibits significant improvement compared to the ACONV system at low signal-to-noise ratio (SNR), and it is shown that, for an M=4 hypotheses, SNR=-10 dB, and a desired P(e)=10(-2), the AFSI system requires 30 times fewer measurements than the ACONV system. Experimental results validating the AFSI system are presented.

Recognition using information-optimal adaptive feature-specific imaging

We present an information-theoretic adaptive feature-specific imaging (AFSI) system for a M-class recognition task. The proposed system utilizes the recently developed task-specific information (TSI) framework to incorporate the knowledge from previous measurements and adapt the projection matrix at each step. The decision-making framework is based on sequential hypothesis testing. We quantify the number of measurements required to achieve a specified probability of misclassification (P(e)), and we compare the performances of three approaches: the new TSI-based AFSI system, a previously reported statistical AFSI system, and static FSI (SFSI). The TSI-based AFSI system exhibits significant improvement compared with SFSI and statistical AFSI at low signal-to-noise ratio (SNR). It is shown that for M=4 hypotheses, SNR=-20 dB and desired P(e)=10(-2), TSI-based AFSI requires 3 times fewer measurements than statistical AFSI, and 16 times fewer measurements than SFSI. We also describe an extension of the proposed method that is suitable for recognition in the presence of nuisance parameters such as illumination conditions and target orientations.