Catálogo de publicaciones - libros

Dictionary-based compression methods are a popular form of data file compression. LZ77, LZ78 and their variants are likely the most famous of these methods. These methods are implemented to reduce the one-dimensional correlation in data, since they are designed to compress text. Therefore, they do not take advantage of the fact that, in images, adjacent pixels are correlated in two dimensions. Previous attempts have been made to linearize images in order to make them suitable for dictionary-based compression, but results show that no single linearization is best for all images. In this paper, a two-dimensional dictionary-based lossless image compression scheme for grayscale images is introduced. Testing results show that the compression performance of the proposed scheme outperforms and surpasses any other existing dictionary-based compression scheme. The results also show that it slightly outperforms JPEG-2000’s compression performance, when it operates in its lossless mode, and it is comparable to JPEG-LS’s compression performance, where JPEG-2000 and JPEG-LS are the current image compression standards.

Image communication is a significant research area which involves improvement in image coding and communication techniques. In this paper, Principal Component Analysis (PCA) is used for face image coding and the coded images are protected with convolutional codes for transmission over Additive White Gaussian Noise (AWGN) channel. Binary Phase Shift Keying (BPSK) is used for the modulation of digital (binarized) coded images. Received binarized coded images are first decoded by the convolutional decoder using the Viterbi algorithm and then PCA decoded for recognition of the face. Unequal error protection (UEP) with two convolutional encoders with different rates is used to increase the overall performance of the system. The recognition rate of the transmitted coded face images without any protection is 35%, while equal protection with convolutional codes gives rates up to 85% accuracy. On the other hand, the proposed UEP scheme provides recognition rates up to 95% with reduced redundancy.

For a hardware random number generator (RNG) in a crypto module, it is important that the RNG hardware offers an output bit stream that is always unbiased. J. H., et al. proposed a combination of the hardware component and a software filter algorithm. However, even though the hardware generating processor generates an output bit stream quickly, if the software filter algorithm is inefficient, the RNG becomes time consuming, thereby restricting the conditions when an RNG can be applied. Accordingly, this paper proposes an effective method of software filtering for an RNG processor in a crypto module. To consistently guarantee the randomness of the output sequence from a RNG, the origin must be stabilized, regardless of any change in circumstances. Therefore, a tree model is proposed to apply the filter algorithm, making it less time consuming than J. H.’s conventional filter algorithm scheme.

A novel multiple description coding framework, called Layered Multiple Description Packetized Coding (LMDPC) is proposed in this paper. We first develop a two description coding scheme from SNR scalable layer coding, where the base layer is duplicated into both descriptions and enhancement layer was split into two parts and sent to separate descriptions, respectively. Two descriptions are then partitioned horizontally and vertically, forming row packets and column packets for transportation. Because each row packet and column packet has only limited intersection, even packets lost happen on both descriptions, the proposed algorithm still has very good error resilient ability. Experimental results have verified the performance of the proposed MDC framework.

SP(Synchronization-Predictive) frame coding, which enables high efficiency of switching between two video bitstreams with different qualities, is supported by H.264/AVC. And FGS(Fine-Granular-Scalability) coding is supported by MPEG-4 video standard. This paper proposes a solution for combination these two tools with each other so as to adapt to high bandwidth variations of Internet or wireless networks and to low bandwidth variations flexibly for transmitted video streams. Experimental results show that our proposed system outperforms FGS coding by 0.47dB and the H.264/AVC-based video stream switching approach by 0.23dB on average.

Imagine is a stream-based prototype processor designed for media processing. It uses a three-level bandwidth hierarchy to exploit parallelism and data locality. It has good performance in media processing. H.264 is the newest digital video coding standard. It can achieve high coding efficiency at the cost of complex computation. In addition, video pictures have natural stream features, such as good special locality and limited temporal dependency. This paper presents an accelerated implementation of motion estimation, which is the most time-consuming part in H.264 coding framework, on Imagine stream processor. Experimental results show that the coding efficiency for QCIF format can be up to 372fps and surpass real-time requirement. The acceleration of stream processing is significant. It proves that H.264 coding is suited for implementation on Imagine.

MPEG-2 test stream for evaluation the static picture quality of digital television (DTV) should meet both good picture quality and stable bit rate. In this paper, we present a method for generating a high quality test stream to evaluate the static picture quality in DTV receiver. The proposed method is suitable for encoding the static test pattern, such as multiburst and crosshatch, and is based on user-defined quantization and adaptive zero stuffing algorithm. The user-defined quantization is suitable for minimizing the quantization error, which is the reason of degradation of picture quality, and the adaptive zero stuffing algorithm is used to solve the overflow of video buffer verifier (VBV) buffer while encoding process by MPEG-2 encoder. Experimental results show that the average PSNR and the bit rate of the proposed method have more efficient and stable than those of the conventional.

MPEG-4 visual simple profile is a widely used video compression standard for mobile solutions. In general, MPEG-4 video decoder requires high computation power for its complex algorithms. It’s difficult to implement MPEG-4 video decoder on hand-held devices directly. In this paper, we proposed a novel color space transform algorithm and optimized the memory access operations. Moreover, the multiperless integer IDCT is adopted to further speed up the decoder. Our optimization is based on ARM7TDMI and ARM920T, which are very desirable cores to mobile solutions for low power consumption. Experimental results show that the optimized decoder acts about 5 times faster than existing XVID MPEG-4 video decoder with small video quality degradation and supports real-time video applications.

Efficient matching methods are crucial in Image Processing. In the present paper we outline a novel algorithm of ”stable marriages” that is also fair and globally satisfactory for both populations to be paired. Our applicative examples here being stereo or motion we match primitives based on level lines segments, known for their robustness to contrast changes. They are separately extracted from images, and we draft the corresponding process too. Then for marriages to be organised each primitive needs to be given a preference list sorting potential mates in the antagonist image: parameters of the resemblance founding preferences are explained. Eventually all operators above are embedded within a recursive least squares method and results are shown and compared with a successful Hough based matching that we had used so far.

The purpose of this paper is to allow for high level shape representation and matching in multi-object images by detecting and extracting the envelope of object groupings in the image. The proposed algorithm uses hierarchical clustering to find object groupings based on spatial proximity as well as low-level shape features of objects in the image. Each grouping is then merged using a morphological algorithm. The envelope is extracted by reconstructing the object from its dynamically pruned concavity tree. We test our approach on a set of 45 multi-object trademark images and we report results on object groupings and envelope extraction.