Heesung Kwon

Kernel-Based Anomaly Detection in Hyperspectral Imagery

Abstract : In this paper we present a nonlinear version of the wellknown anomaly detection method referred to as the RXalgorithm. Extending this algorithm to a feature space associated with the original input space via a certain nonlinear mapping function can provide a nonlinear version of the RX-algorithm. This nonlinear RX-algorithm, referred to as the kernel RX-algorithm, is basically intractable mainly due to the high dimensionality of the feature space produced by the non-linear mapping function. However, in this paper it is shown that the kernel RX-algorithm can easily be implemented by kernelizing it in terms of kernels which implicitly compute dot products in the feature space. Improved performance of the kernel RX-algorithm over the conventional RX-algorithm is shown by testing several hyperspectral imagery for military target and mine detection.

Compression of SAR imagery using adaptive residual vector quantization

Compression of SAR imagery for battlefield digitization is discussed in this paper. THe images are first processed to separate out possible target areas. These target areas are compressed losslessly to avoid any degradation of the images. The background information which is usually necessary to establish context, is compressed using a hybrid vector quantization algorithm. An adaptive variable rate residual vector quantizer is use to compress the residual signal generated by a neural network predictor. The vector quantizer codebooks are optimized for entropy coding using an entropy-constrained algorithm to further improve the coding performance. This constrained vector-quantizer combination performs extremely well as suggested by the experimental results.

Segmentation based wavelet coding of digital images

In this paper, we present a segmentation based wavelet coding scheme, in which an image is segmented into two regions: stationary areas (background) and the areas containing edge information (foreground). These regions are then encoded independently using two dedicated encoders that are optimized for each region. A 2-D edge operator is used for segmenting the image. We use the embedded zerotree wavelet (EZW) algorithm for encoding the background due to its good performance on stationary areas. The foreground area is, however, encoded using a predictive residual vector quantizer (PRVQ). Experimental results show that the proposed technique improves the quality of the reconstructed images, both numerically (in terms of mean square error) and perceptually when compared to EZW at the same bit rate.

High compression of SAR imagery for battlefield surveillance

In this paper a compression algorithm is developed to compress SAR imagery at very low bit rate. A new vector quantization (VQ) technique called the predictive residual vector quantizer (PRVQ) is presented for encoding the SAR imagery. Also a variable-rate VQ scheme called the entropy- constrained PRVQ (EC-PRVQ), which is designed by imposing a constraint on the output entropy of the PRVQ, is designed. Experimental results are presented for both PRVQ and EC-PRVQ at high compression ratios. The encoded images are also compared with that of a wavelet-based coder.

Target Classification Using Adaptive Feature Extraction and Subspace Projection for Hyperspectral Imagery

Hyperspectral imaging sensors have been widely studied for automatic target recognition (ATR), mainly because a wealth of spectral information can be obtained through a large number of narrow contiguous spectral channels (often over a hundred). Targets are man-made objects (e.g., vehicles) whose constituent materials and internal structures are usually substantially different from natural objects (i.e., backgrounds). The basic premise of hyperspectral target classification is that the spectral signatures of target materials are measurably different than background materials, and most approaches further assume that each relevant material, characterized by its own distinctive spectral reflectance or emission, can be identified among a group of materials based on spectral analysis of the hyperspectral data.

Hyperspectral Target Detection Based on Kernels

In this chapter, linear signal or target detection algorithms are extended to nonlinear versions by using kernel-based methods. In kernel-based methods, learning is implicitly performed in a high-dimensional feature space where high order correlation or nonlinearity within the data are exploited. Nonlinear realization is mainly pursued to reduce data complexity in a high-dimensional feature space and consequently provide simpler decision rules for data discrimination.

Very-low-bit-rate video coding using quadtree decomposition and cache-based vector quantization

We investigate the use of vector quantizers (VQs) with memory to encode image sequences. A video coding technique using cache-based vector quantization is presented. In this technique, a small codebook (subcodebook), meaning a cache memory codebook, is used to encode the input signals. This small subcodebook is dynamically generated from a much larger codebook (supercodebook) through a caching scheme. Therefore, the subcodebook dynamically adapts to the local characteristics of the motion-compensated residual signal, which varies with time. Both least recently used (LRU) and modified LRU cache-replacement techniques are used. An efficient bit-allocation strategy using quadtree decomposition is used with the cache-based VQ to compress the video signal. The proposed video codec outperforms H.263 in terms of PSNR and perceptual quality at very low bit rates, ranging from 5 to 15 kbps. Experimental results are presented for two image sequences, salesman and Miss America.

Multiscale video compression using adaptive finite-state vector quantization

We investigate the use of vector quantizers (VQs) with memory to encode image sequences. A multiscale video coding technique using adaptive finite-state vector quantization (FSVQ) is presented. In this technique, a small codebook (subcodebook) is generated for each input vector from a much larger codebook (supercodebook) by the selection (through a reordering procedure) of a set of appropriate codevectors that is the best representative of the input vector. Therefore, the subcodebook dynamically adapts to the characteristics of the motion-compensated frame difference signal. Several reordering procedures are introduced, and their performance is evaluated. In adaptive FSVQ, two different methods, predefined thresholding and rate-distortion cost optimization, are used to decide between the supercodebook and subcodebook for encoding a given input vector. A cache-based vector quantizer, a form of adaptive FSVQ, is also presented for very-low-bit-rate video coding. An efficient bit-allocation strategy using quadtree decomposition is used with the cache-based VQ to compress the video signal. The proposed video codec outperforms H.263 in terms of the peak signal-to-noise ratio and perceptual quality at very low bit rates, ranging from 5 to 20 kbps. The picture quality of the proposed video codec is a significant improvement over previous codecs, in terms of annoying distortions (blocking artifacts and mosquito noises), and is comparable to that of recently developed wavelet-based video codecs. This similarity in picture quality can be explained by the fact that the proposed video codec uses multiscale segmentation and subsequent variable-rate coding, which are conceptually similar to wavelet-based coding techniques. The simplicity of the encoder and decoder of the proposed codec makes it more suitable than waveletbased coding for real-time, very-low-bit-rate video applications.

Very low bit rate compression for one-to-N video broadcast

Compression of video for low bitrate communication is studied in this paper. Use of vector quantization in the H.263 framework is proposed. A variable rate residual vector quantizer with a transform vector quantizer in the first stage is used along with a strategy to adapt the bit-rate to the activity in the block. This ability to adapt the bit- rate is very important for very low bitrate compression. The proposed multistage quantizer combined with an adaptive arithmetic codec produced very good results. The variability in the bit-rate was achieved by using smaller block sizes in the later stages of quantizer along with selective quantization of only high energy blocks at the later stages. Performance comparison of the proposed codec with that of H- 263 indicates that there is superior compression results especially bitrates less than 8kb/s.