Minh Dao

Structured Sparse Priors for Image Classification

Model-based compressive sensing (CS) exploits the structure inherent in sparse signals for the design of better signal recovery algorithms. This information about structure is often captured in the form of a prior on the sparse coefficients, with the Laplacian being the most common such choice (leading to l1-norm minimization). Recent work has exploited the discriminative capability of sparse representations for image classification by employing class-specific dictionaries in the CS framework. Our contribution is a logical extension of these ideas into structured sparsity for classification. We introduce the notion of discriminative class-specific priors in conjunction with class specific dictionaries, specifically the spike-and-slab prior widely applied in Bayesian sparse regression. Significantly, the proposed framework takes the burden off the demand for abundant training image samples necessary for the success of sparsity-based classification schemes. We demonstrate this practical benefit of our approach in important applications, such as face recognition and object categorization.

Multi-task image classification via collaborative, hierarchical spike-and-slab priors

Promising results have been achieved in image classification problems by exploiting the discriminative power of sparse representations for classification (SRC). Recently, it has been shown that the use of class-specific spike-and-slab priors in conjunction with the class-specific dictionaries from SRC is particularly effective in low training scenarios. As a logical extension, we build on this framework for multitask scenarios, wherein multiple representations of the same physical phenomena are available. We experimentally demonstrate the benefits of mining joint information from different camera views for multi-view face recognition.

Hierarchical sparse modeling using Spike and Slab priors

Sparse modeling has demonstrated its superior performances in many applications. Compared to optimization based approaches, Bayesian sparse modeling generally provides a more sparse result with a knowledge of confidence. Using the Spike and Slab priors, we propose the hierarchical sparse models for the scenario of single task and multitask - Hi-BCS and CHi-BCS. We draw the connections of these two methods to their optimization based counterparts and use expectation propagation for inference. The experiment results using synthetic and real data demonstrate that the performance of Hi-BCS and Chi-BCS are comparable or better than their optimization based counterparts.

Collaborative Multi-Sensor Classification Via Sparsity-Based Representation

In this paper, we propose a general collaborative sparse representation framework for multi-sensor classification, which takes into account the correlations as well as complementary information between heterogeneous sensors simultaneously while considering joint sparsity within each sensors observations. We also robustify our models to deal with the presence of sparse noise and low-rank interference signals. Specifically, we demonstrate that incorporating the noise or interference signal as a low-rank component in our models is essential in a multi-sensor classification problem when multiple co-located sources/sensors simultaneously record the same physical event. We further extend our frameworks to kernelized models which rely on sparsely representing a test sample in terms of all the training samples in a feature space induced by a kernel function. A fast and efficient algorithm based on alternative direction method is proposed where its convergence to an optimal solution is guaranteed. Extensive experiments are conducted on several real multi-sensor data sets and results are compared with the conventional classifiers to verify the effectiveness of the proposed methods.

Group structured dirty dictionary learning for classification

Dictionary learning techniques have gained tremendous success in many classification problems. Inspired by the dirty model for multi-task regression problems, we proposed a novel method called group-structured dirty dictionary learning (GDDL) that incorporates the group structure (for each task) with the dirty model (across tasks) in the dictionary training process. Its benefits are two-fold: 1) the group structure enforces implicitly the label consistency needed between dictionary atoms and training data for classification; and 2) for each class, the dirty model separates the sparse coefficients into ones with shared support and unique support, with the first set being more discriminative. We use proximal operators and block coordinate decent to solve the optimization problem. GDDL has been shown to give state-of-art result on both synthetic simulation and two face recognition datasets.

Real Time Compressive Sensing Video Reconstruction in Hardware

Compressive sensing has allowed for reconstruction of missing pixels in incomplete images with higher accuracy than was previously possible. Moreover, video data or sequences of images contain even more correlation, leading to a much sparser representation as demonstrated repeatedly in numerous digital video formats and international standards. Compressive sensing has inspired the design of a number of imagers which take advantage of the need to only subsample a scene, which reduces power consumption by requiring acquisition and transmission of fewer samples. In this paper, we show how missing pixels in a video sequence can be estimated using compressive sensing techniques. We present a real time implementation of our algorithm and show its application to an asynchronous time-based image sensor (ATIS) from the Austrian Institute of Technology. The ATIS only provides pixel intensity data when and where a change in pixel intensity is detected, however, noise randomly causes intensity changes to be falsely detected, thereby providing random samples of static regions of the scene. Unlike other compressive sensing imagers, which typically have pseudo-random sampling designed in at extra effort, the ATIS used here provides random samples as a side effect of circuit noise. Here, we describe and analyze a field-programmable gate array implementation of a matching pursuit (MP) algorithm for compressive sensing reconstruction capable of reconstructing over 1.9 million 8 × 8 pixel regions per second with a sparsity of 11 using a basis dictionary containing 64 elements. In our application to ATIS we achieve throughput of 28 frames per second at a resolution of 304 × 240 pixels with reconstruction accuracy comparable to that of state of the art algorithms evaluated offline.

Temporal rate up-conversion of synthetic aperture radar via low-rank matrix recovery

The radar data to form synthetic aperture radar (SAR) imagery is normally transmitted and received by moving platforms like aircraft or vehicles. In many situations, the platforms move at high speed; which reduces the number of sampling records collected to the synthetic aperture, hence degrades the quality of the reconstructed SAR images. Therefore, it is necessary to develop an algorithm that is capable of increasing the temporal frequency rate of the received data. In this paper, we propose a novel technique to generate intermediate records from the existing ones by a locally-adaptive low-rank matrix recovery framework. The system first fills in the blank records using a bi-directional motion estimation scheme. The initialized aperture records are then refined by a robust low-rank matrix completion algorithm using the reference from neighborhood clean records. Experiments demonstrate that the proposed method provides comparative results when up-converting the aperture rate by a factor of two or four, both in mean square error of the raw SAR signal and PSNR performance of the recovered SAR images.

Structured sparse representation with low-rank interference

This paper proposes a novel framework that is capable of extracting the low-rank interference while simultaneously promoting sparsity-based representation of multiple correlated signals. The proposed model provides an efficient approach for the representation of multiple measurements where the underlying signals exhibit a structured sparsity representation over some proper dictionaries but the set of testing samples are corrupted by the interference from external sources. Under the assumption that the interference component forms a low-rank structure, the proposed algorithms minimize the nuclear norm of the interference to exclude it from the representation of multivariate sparse representation. An efficient algorithm based on alternating direction method of multipliers is proposed for the general framework. Extensive experimental results are conducted on two practical applications: chemical plume detection and classification in hyperspectral sequences and robust speech recognition in noisy environments to verify the effectiveness of the proposed methods.

Chemical plume detection in hyperspectral imagery via joint sparse representation

In this paper, we propose a new spatial-temporal joint sparsity method for the identification and detection of chemical plume in hyperspectral imagery. The proposed algorithm relies on two key observations: 1. each hyperspectral pixel can be approximately represented by a sparse linear combination of the training samples; and 2. neighborhood pixels from the same hyperspectral image as well as consecutive hyperspectral frames usually have similar spectral characteristics. By grouping these pixels into a joint group structure and forcing them to have the same sparsity support of the training samples, we effectively exclude the correlation of not only spatial but also time domain of the HSI data. Before the presence of this paper, almost no methods have made use of the temporal information for the detection of chemical plume in hyperspectral video data. Furthermore, the proposed method shows very competitive results with the Adaptive Matched Subspace Detector (AMSD) algorithm where the chemical types are predefined.

Video concealment via matrix completion at high missing rates

Video error concealment is an important technique in video communication to recover corrupted parts when erroneously transmitting compressed sequences over network. In this work, we propose a novel error concealment scheme by grouping similar patches in the temporal domain to construct a low-rank matrix and recover the missing areas by the matrix completion technique. Different from most state-of-the-art algorithms which recover the lost blocks based on at least one clean frame (I-frame), the proposed algorithm can work for the most general case when all frames are violated. When having I-frames in the receiver, the proposed algorithm also achieves much higher PSNR compared with the Boundary Matching Algorithm (BMA) adopted in H.264.