Catálogo de publicaciones - libros

Topographical images such as aerial and satellite images are usually similar with respect to colors and textures but not in shapes. Thus shape features of the images and the methods of extracting them become critical for effective image retrieval from topographical image databases. In this paper, we propose a shape feature extraction method as well as a similarity function for topographical image retrieval. The method extracts a set of attributes which can model the presence of holes and disconnected regions in images and is tolerant to pre-processing, more specifically segmentation, errors. Experiments suggest that retrieval using attributes extracted using the proposed method performs better than using the Curvature Scale Space (CSS) which is a well known method in the area of shape feature extraction.

In this paper we propose a robust text segmentation method in complex background. The proposed method first utilizes the K-means algorithm to decompose a detected text block into different binary image layers. Then an effective post-processing is followed to eliminate background residues in each layer. In this step we develop a group of robust constraints to characterize general text regions based on color, edge and stroke thickness. We also propose the components relation constraint (CRC) designed specifically for Chinese characters. Finally the text image layer is identified based on the periodical and symmetrical layout of text lines. The experimental results show that our method can effectively eliminate a wide range of background residues, and has a better performance than the K-means method, as well as a high speed.

In projection display systems, specular reflection may distract users and obscure the projected useful information. This paper demonstrates that the specular reflection can be avoided by redundantly illuminating the display surface using multiple overlapping projectors and cameras. Our initial system using two projectors and a single camera is presented. The system automatically estimates where specular reflection occurs and which projector generates the specular reflection. Then, the system blanks the light of the projector falling on the specular region while boosting the other projector’s light so that the additional light is projected onto the surface consistently. Through experiments, it is shown that the system eliminates specular reflection successfully while maintaining a good image quality.

Image classification could be treated as an effective solution to enable keyword-based semantic image retrieval. In this paper, we propose a novel image classification framework by learning semantic concepts of image categories. To choose representative features for an image category and meanwhile reduce noisy features, a three-step salient feature selection strategy is proposed. In the feature selection stage, salient patches are first detected and clustered. Then the region of dominance and salient entropy measures are calculated to reduce non-common salient patches for the category. Based on the selected visual keywords, SVM and keyword frequency model categorization method are applied to classification, respectively. The experimental results on Corel image database demonstrate that the proposed salient feature selection approach is very effective in image classification and visual concept learning.

Contourlet is a new image representation method, which can efficiently represent contours and textures in images. In this paper, we analyze the distribution of significant contourlet coefficients in different subbands and propose a contourlet image coding algorithm by constructing a virtual low frequency subband and adjusting coding method of SPIHT (Set Partitioning in Hierarchical Trees) algorithm according to the structure of contourlet coefficients. The proposed coding algorithm can provide an embedded bit stream, which is very desirable in heterogeneous networks. Our experiments demonstrate that the proposed coding algorithm can achieve better or competitive compression performance when compared with traditional wavelet transform with SPIHT and wavelet-based contourlet transform with SPIHT, which both are embedded image coding algorithms based on two non-redundant transforms. At the same time, benefiting from genuine contourlet adopted in the proposed coding algorithm, more contours and textures in the coded images are preserved to ensure superior subjective quality.

In this paper, attention is paid to the automatic generation of XML-based descriptions containing information about the high-level structure of binary multimedia resources. These structural metadata can then be transformed in order to reflect a desired adaptation of a multimedia resource, and can subsequently be used to create a tailored version of the resource in question. Based on this concept, two technologies are presented: MPEG-21 BSDL and a modified version of XFlavor being able to create BSDL compatible output. Their usage is elaborated in more detail with respect to the valid exploitation of multi-layered temporal scalability in H.264/MPEG-4 AVC’s base specification, and in particular with a focus on a combined usage of the sub-sequence coding technique and Supplemental Enhancement Information (SEI) messages. Some performance measurements in terms of file sizes and computational times are presented as well.

The primary goal of this paper is to facilitate the rate-distortion control in compressed domain, without introducing a full decoding and re-encoding system in pixel domain. For this aim, the error propagation behavior over several frame-sequences due to DCT coefficients-drop is investigated on the basis of statistical and empirical properties. Then, such properties are used to develop a simple estimation model for the CD distortion accounting for the characteristics of the underlying coded-frame. Experimental results show that the proposed model allows us to effectively control rate-distortions into coded-frames over different kinds of video sequences.

We propose an error concealment technique to recover lost motion vectors at the H.264 decoder side. To recover the lost motion vectors, there are two simple techniques: (1) no prediction, where the lost motion vectors are set to zeros, and (2) the prediction using the average or median of spatially adjacent blocks’ motion vectors [1]. In this paper, we propose a Kalman filter based scheme for motion vector recovery, and experimented with two test image sequences: Mobile&Calendar and Susie. The experimental results show that our Kalman filter based motion vector recovery scheme improves at average 2 dB PSNR over conventional H.264 decoding with no error recovery. We also improve our scheme using Hilbert curve scan order for Kalman input, and we get 0.512 – 1.652 dB PSNR improvements with better subjective quality over line-by-line scan order.

Entropy coding is one of the most important techniques in video codec. Two main criteria to assess an entropy coder are coding efficiency and friendly realization characteristic. In the recent development of the H.264/AVC standard, a sophisticated entropy coder, named Context-based Adaptive Variable Length Coding (CAVLC), has been invented, which supplies higher coding efficiency than other VLC-based entropy coders in previous video coding standards. But due to its algorithm’s inherit traits CAVLC must be executed in a sequential manner, which results in relative low throughput rate. In this paper, a new well-designed context-based VLC entropy coder for transform coefficients is presented. It exploits coefficient block’s inner context information to obtain high coding efficiency, and at the same time context models for successive coding elements are designed to be dependent-free so that the coder is more apt to be parallel. Experimental results show that the proposed entropy coder can exhibit the same coding efficiency as CAVLC. Therefore, a new high performance entropy coder with characteristics of parallel orientation and high coding efficiency is supplied.

Previous works related to texture coordinate coding of the three-dimensional(3-D) mesh models employed the same predictor as the geometry coder. However, discontinuities in the texture coordinates cause unreasonable prediction. Especially, discontinuities become more serious for the 3-D mesh model with a non-atlas texture image. In this paper, we propose a new coding scheme to remove discontinuities in the texture coordinates by reallocating texture segments according to a coding order. Experiment results show that the proposed coding scheme outperforms the MPEG-4 3DMC standard in terms of compression efficiency. The proposed scheme not only overcome the discontinuity problem by regenerating a texture image, but also improve coding efficiency of texture coordinate compression.