Oligoquantization in low-rate lossy source coding

The theory of high rate lossy source coding is well developed both with respect to the practice of quantization and its fundamental rate distortion limits. But many modern compression systems for natural signals such as audio and images operate at lower rates, which are not covered by these theories. Often these systems are based on a signal transform followed by a nonlinear processing step that reduces the number of coefficients that are actually quantized. Both this nonlinearity and the non-Gaussianity of the involved signals rule out using the theory of (high rate) linear transform coding to analyze such compression algorithms. The lossy source coding theorem could be used, but it is not the right tool "to see what happens". Another way of getting a better understanding of these systems is to look at them as a form of nonlinear approximation of individual signals. While partly avoiding the need for statistical signal models that are hard to come by, this approach still relies on high rate results for the coefficient quantization. In this thesis we take a more information theoretic approach, whose main result is a new low-rate upper bound on the distortion rate function. In order to obtain that result, the problem had to be simplified substantially. To begin with, only i.i.d. memoryless sources with a mean square error distortion measure are considered. Instead of bounding the distortion rate function D(R) directly, we derive upper bounds on the optimum performance of a certain class of quantizers, which by definition upper bound also D(R). Inspired by the observation that many of those compression systems quantize only very few coefficients at low rates, we called this the class of oligoquantizers. In particular, if the coefficients have a peaked unimodal probability density, a simple magnitude thresholding operation is very effective at deciding which coefficients should be quantized. It turns out that peaked unimodal densities are also the most interesting ones, not just because they appear in many transform coding systems, but also because they have a D(R) behavior that is very different from a Gaussian. Thus the emphasis of this work will be on oligoquantization based on scalar thresholding. Actually we will show that the principle of oligoquantization does not generalize well to higher dimensions. On a more applied note, we analyze several models for sparse transform coefficients with the aid of the low rate bound. For the special case of sparse coefficient vectors and Hamming distortion measure, analytic expressions for R(D) are obtained. A high rate bound that complements its low rate counterpart is also derived. Finally, we show how these bounds can be computed directly from a set of source samples, without first estimating the underlying probability density. These empirical bounds can be a useful complement to the rate distortion function computed with Blahut's algorithm, since the latter is hard to use at very low or high rates.

Polar codes for channel and source coding

Satish Babu Korada

The two central topics of information theory are the compression and the transmission of data. Shannon, in his seminal work, formalized both these problems and determined their fundamental limits. Since then the main goal of coding theory has been to find practical schemes that approach these limits. Polar codes, recently invented by Arikan, are the first "practical" codes that are known to achieve the capacity for a large class of channels. Their code construction is based on a phenomenon called "channel polarization". The encoding as well as the decoding operation of polar codes can be implemented with O(N log N) complexity, where N is the blocklength of the code. We show that polar codes are suitable not only for channel coding but also achieve optimal performance for several other important problems in information theory. The first problem we consider is lossy source compression. We construct polar codes that asymptotically approach Shannon's rate-distortion bound for a large class of sources. We achieve this performance by designing polar codes according to the "test channel", which naturally appears in Shannon's formulation of the rate-distortion function. The encoding operation combines the successive cancellation algorithm of Arikan with a crucial new ingredient called "randomized rounding". As for channel coding, both the encoding as well as the decoding operation can be implemented with O(N log N) complexity. This is the first known "practical" scheme that approaches the optimal rate-distortion trade-off. We also construct polar codes that achieve the optimal performance for the Wyner-Ziv and the Gelfand-Pinsker problems. Both these problems can be tackled using "nested" codes and polar codes are naturally suited for this purpose. We further show that polar codes achieve the capacity of asymmetric channels, multi-terminal scenarios like multiple access channels, and degraded broadcast channels. For each of these problems, our constructions are the first known "practical" schemes that approach the optimal performance. The original polar codes of Arikan achieve a block error probability decaying exponentially in the square root of the block length. For source coding, the gap between the achieved distortion and the limiting distortion also vanishes exponentially in the square root of the blocklength. We explore other polar-like code constructions with better rates of decay. With this generalization, we show that close to exponential decays can be obtained for both channel and source coding. The new constructions mimic the recursive construction of Arikan and, hence, they inherit the same encoding and decoding complexity. We also propose algorithms based on message-passing to improve the finite length performance of polar codes. In the final two chapters of this thesis we address two important problems in graphical models related to communications. The first problem is in the area of low-density parity-check codes (LDPC). For practical lengths, LDPC codes using message-passing decoding are still the codes to beat. The current analysis, using density evolution, evaluates the performance of these algorithms on a tree. The tree assumption corresponds to using an infinite length code. But in practice, the codes are of finite length. We analyze the message-passing algorithms for this scenario. The absence of tree assumption introduces correlations between various messages. We show that despite this correlation, the prediction of the tree analysis is accurate. The second problem we consider is related to code division multiple access (CDMA) communication using random spreading. The current analysis mainly focuses on the information theoretic limits, i.e., using Gaussian input distribution. However in practice we use modulation schemes like binary phase-shift keying (BPSK), which is far from being Gaussian. The effects of the modulation scheme cannot be analyzed using traditional tools which are based on spectrum of large random matrices. We follow a new approach using tools developed for random spin systems in statistical mechanics. We prove a tight upper bound on the capacity of the system when the user input is BPSK. We also show that the capacity depends only on the power of the spreading sequences and is independent of their exact distribution.

EPFL2009

Image security and processing in the JPEG 2000 compressed domain

Raphaël Grosbois

Digital images are becoming increasingly successful thanks to the development and the facilitated access to systems permitting their generation (i.e. camera, scanner, imaging software, etc). A digital image basically corresponds to a 2D discrete set of regularly spaced samples, called pixels, where each pixel contains the light intensity information (e.g., luminance, chrominance) of a very localized spatial region of the image. In the case of natural images, pixel values are acquired through one or several arrays of MOS semiconductors (Charge Couple Devices, CCDs), each generating an electrical information proportional to the incoming light intensity. The initial finalities of digital images were the storage on a dedicated medium (e.g., camera's memory, computer's hard drive, CDROM), eventual transmissions, and final display on a screen or printing. With such a narrow scope, the principal goal of image processing and coding tools was to face storage and transmission bandwidth limitations thanks to efficient compression algorithms reducing the image representation size. However, with recent developments in computing, algorithmic and telecommunication domains, many new applications (i.e. web-publishing, remote browsing etc.) have arisen. They generally require additional and enhanced features (i.e. progressive decoding, random-access, region of interest support, robustness to transmission errors, etc.) and have motivated the creation of a new generation of coding algorithms which, besides their good compression performance, present many other useful features. Hence, digital images are almost never represented as a simple set of pixel values (i.e. raw representation) but, instead, under a specific compact way (i.e. compressed or coded representation), chosen according to features it brings to the considered application. A compressed version of an image is obtained by removing as much spatial, visual and statistical redundancies as possible, thanks to appropriate coding methods, while keeping an acceptable visual quality. Noting that natural images have most of their energy concentrated in low frequency components, recent coding algorithms generally first decompose the image into a specific frequency domain DCT, DWT, etc). The goal is to obtain a representation, where few coefficients are sufficient for reconstructing the image with a good quality. The precision of transformed coefficients is then generally reduced by quantization in order to make them more compressible by an entropy coder, aiming at removing statistical redundancies of quantization indexes. The ultimate compressed representation, called codestream, is usually obtained by a rate-allocation process that tries to achieve the best trade-off between the compression ratio and the reconstructed image quality. JPEG 2000, the new still image coding standard developed by the Joint Photographic Experts Group JPEG), is based on these state-of-the-art compression techniques, but is also designed to fulfill many requirements of recent applications. It only normalizes the decoding algorithm and consequently let some liberty for designing encoders optimized for some specific features or for building extensions that take profit from its compressed domain specifics. The success of the JPEG 2000 standard will not only depend on its intrinsic performance, but also on its ability to comply with specific demands of actual and future imaging applications. Thus, among the major concerns of image content providers are the security of the transmission over networks and of the image itself: with current facilities to instantly access on-line digital libraries from anywhere in the world, to perfectly copy and to easily modify the content of an image, solutions must be found in order to permit, at one hand, Intellectual Property Right (IPR) protection and image integrity verification and, at the other hand, the development of dedicated tools favoring exchange and purchase of images over communication networks. It is worth pointing out that some cryptographic-based solutions to these problems already exist. However, they need to be adapted to the digital image and its compressed domain representation in order to take full advantage from their specifics and to avoid restraining fields of applications. The JPEG 2000 coding algorithm is mainly based on Discrete Wavelet Transform (DWT), embedded scalar quantization and adaptive arithmetic coding. From a terminology point of view, this means that compressed domain representation can indifferently refer to either wavelet coefficients, or quantized wavelet coefficients, or bit streams (i.e. entropy coded group of quantization indexes) or the codestream (i.e. aggregation of bit streams and headers containing necessary decoding information). The choice of the appropriate compressed domain actually depends on the considered application. The quality of a JPEG 2000 image, at a given compression ratio, mainly depends on the rate-allocation procedure used at the encoder side. Such a procedure applies on entropy coded quantization indexes and favors groups of quantization indexes (i.e. code-blocks) offering the best rate-distortion trade-offs. However, this does not necessary correspond to the most interesting part of the image, from an end-observer point of view. Hence, the standard provides with ways to define Regions Of Interest (ROI) which are prioritized during the encoding process in order to exhibit a higher quality than the rest (i.e. background) at any decoding time. This feature applies either in the quantized wavelet domain or at the bit stream level, but its parameters are generally not very explicit for a standard end-user and only provide with a rough control of the decoded ROI quality. Consequently, the first objective of this thesis is to create and also extend compressed domain tools for controlling the quality of a JPEG 2000 ROI. In the meantime, several image processing techniques, such as watermarking, are applied directly in the spatial domain or in a specific transform space defined from the spatial domain. However, since digital images are preferably available under a compressed/encoded format, which we assume to be JPEG 2000 in this thesis, their implementation first imply decompressing the image, then apply the considered processing task and finally re-encode the resulting image. Such a scheme has two main drawbacks: First, it generally implies time and complexity overheads compared to equivalent methods (if they exist) in the JPEG 2000 compressed domain. Such effects become important when the scheme is repeatedly applied on multiple images. Second, since encoding and decoding operations are generally lossy, the introduced distortion can become non negligible whenever several processing tasks are repeated on a same image. These observations lead to the second objective of this thesis, which is to adapt or create a watermarking algorithm dedicated to the JPEG 2000 compressed domain. Finally, there are imaging algorithms, such as authentication and access control, that already exist and could be directly applied to JPEG 2000 images but, because they do not take into account the coding algorithm specifics, they either decrease compression performance or remove useful features of the encoded representation (scalability, random access, etc). This leads to the third objective of this thesis, which is to adapt, create and combine authentication and access control algorithms with JPEG 2000 coding and decoding. Thus, the common goal of the three objectives described above is the deep integration of selected processing and security algorithms into a JPEG 2000 codec in order to provide with minimum complexity, JPEG 2000 compliant codestreams and an unified framework for many imaging applications.

EPFL2003

Approaching the Rate-Distortion Limit by spatial Coupling with Belief Propagation and Decimation

Vahid Aref, Nicolas Macris, Marc Vuffray

We investigate an encoding scheme for lossy compression based on spatially coupled Low-Density Generator-Matrix codes. The degree distributions are regular, or are Poisson on the code-bit side and check-regular which allows use for any compression rate. The performance of a low complexity Belief Propagation Guided Decimation algorithm is excellent, and for large check degrees it gets close to Shannon's rate-distortion limit. We investigate links between the algorithmic performance and the phase diagram of a relevant random Gibbs measure. The associated dynamical and condensation thresholds are computed within the framework of the cavity method. We observe that: (i) the dynamical threshold of the spatially coupled construction saturates towards the condensation threshold; (ii) for large degrees the condensation threshold approaches the information theoretic test-channel parameter of rate-distortion theory. This provides heuristic insight into the excellent performance of the BPGD algorithm.