Catálogo de publicaciones - libros

This paper develops a novel and efficient dimension reduction scheme–Fast Adaptive Discriminant Analysis (FADA). FADA can find a good projection with adaptation to different sample distributions and discover the classification in the subspace with naïve Bayes classifier. FADA overcomes the high computational cost problem of current Adaptive Discriminant Analysis (ADA) and also alleviates the overfitting problem implicitly caused by ADA. FADA is tested and evaluated using synthetic dataset, COREL dataset and three different face datasets. The experimental results show FADA is more effective and computationally more efficient than ADA for image classification.

Cameras (DSCs) may be a boon to most users whom are simply trigger-happy. However, this automatic mode may not generate the best photos possible or be even applicable for certain types of shots, especially those that require technical expertise. To bridge this gap, many DSCs now offer “Scene Modes” that would easily allow the user to effortlessly configure his camera to specifically take certain types of photos, usually resulting in better quality pictures. These “Scene Modes” provide valuable contextual information about these types of photos and in this paper, we examine how we could make use of “Scene Modes” to assist in generic Image Scene Classification for photos taken on expert/manual settings. Our algorithm could be applied to any image classes associated with the “Scene Modes” and we demonstrated this with the classification of fireworks photos in our case study.

Relevance feedback is an important mechanism for narrowing the semantic gap in content-based image retrieval and the process involves the user labeling positive and negative images. Very often, it is some specific objects or regions in the positive feedback images that the user is really interested in rather than the entire image. This paper presents a hierarchical graphical model for automatically extracting objects and regions that the user is interested in from the positive images which in turn are used to derive features that better reflect the user’s feedback intentions for improving interactive image retrieval. The novel hierarchical graphical model embeds image formation prior, user intention prior and statistical prior in its edges and uses a max-flow/min-cut method to simultaneously segment all positive feedback images into user interested and user uninterested regions. An important innovation of the graphical model is the introduction of a layer of visual appearance prototype nodes to incorporate user intention and form bridges linking similar objects in different images. This architecture not only makes it possible to use all feedback images to obtain more robust user intention prior thus improving the object segmentation results and in turn enhancing the retrieval performance, but also greatly reduces the complexity of the graph and the computational cost. Experimental results are presented to demonstrate the effectiveness of the new method.

Content-based image retrieval (CBIR) has suffered from the lack of linkage between low-level features and high-level semantics. Although relevance feedback (RF) CBIR provides a promising solution involving human interaction, certain query images poorly represented by low-level features still have unsatisfactory retrieval results. An innovative method has been proposed to increase the percentage of relevance of target image database by using graph cuts theory with the maximum-flow/minimum-cut algorithm and relevance feedback. As a result, the database is reformed by keeping relevant images while discarding irrelevant images. The relevance is increased and thus during following RF-CBIR process, previously poorly represented relevant images have higher probability to appear for selection. Better performance and retrieval results can thus be achieved.

There is a great deal of research conducted on hyperplane based query such as Support Vector Machine (SVM) in Content-based Image Retrieval(CBIR). However, the SVM-based CBIR always suffers from the problem of the imbalance of image data. Specifically, the number of negative samples (irrelevant images) is far more than that of the positive ones. To deal with this problem, we propose a new active learning approach to enhance the positive sample set in SVM-based Web image retrieval. In our method, instead of using complex parsing methods to analyze Web pages, two kinds of “lightweight” image features: the URL of the Web image and its visual features, which can be easily obtained, are applied to estimate the probability of the image being a potential positive sample. The experiments conducted on a test data set with more than 10,000 images from about 50 different Web sites demonstrate that compared with traditional methods, our approach improves the retrieval performance significantly.

The AVS video coding standard is established by China. However it is not limited to be a national standard but an open standard. In September 2006, AVS was adopted as a candidate for IPTV standards by the ITU. AVS is becoming more and more well known by the world. Many related products and systems have been released in the past years. This paper will introduce an end-to-end application system of AVS in China—AVS-IPTV. The AVS-IPTV system is built by CNC (China Netcom Group), which is a leading broadband communications and fixed-line telecommunications operator in China. The AVS-IPTV system proves AVS can meet the application requirements well.

This paper unveils the solution of an AVS1-P2 based IPTV (AVS-IPTV) system network trial, which is architected with the common three levels such as national, provincial and urban level, and each system function of each level is described. In order to verify the solution, the AVS-IPTV testing and verification environment has been setup in the China Netcom Group labs, which promoted STB (Set-top box) to support AVS. An AVS-IPTV system network commercial trial has been successfully deployed to test and verify the performance of the AVS-IPTV system function, EPG (Electronic Program Guide), AVS streaming service, and the quality of experience. The trial makes AVS standard and AVS-IPTV system commercially applicable in the real world.

AVS Standard is the abbreviation of Advanced Audio Video Coding Standard made by China with the main purpose to efficiently compress digital audio and video data. The standard may be applied in the field of information industry, such as high-resolution digital broadcast, high-density laser-digital storage media, wireless broadband multimedia communication and broadband stream media. MPEG-2 is the most popular international video compression standard, and has existed in different systems and networks for a long time. At present, most of video programs are made in MPEG-2 format. AVS is expected to become popular in the coming decade. There is a requirement to convert the MPEG-2 programs into AVS ones. This paper presents the usage of MPEG-2 to AVS transcoders in IPTV systems.

Audio and video synchronization is an important technique for many video applications. This paper introduced a new synchronization method used in the emerging AVS system. The decoder system clock is recovered from transport rate info. The audio and video are synchronized according to relative display time info. Some error concealment strategies are also discussed.

Face recognition is one of the most intensively studied topics in computer vision and pattern recognition, but few are focused on how to robustly recognize expressional faces with one single training sample per class. In this paper, we modify the regularization-based optical flow algorithm by imposing constraints on some given point correspondences to obtain precise pixel displacements and intensity variations. By using the optical flow computed for the input expressional face with respect to a referenced neutral face, we remove the expression from the face image by elastic image warping to recognize the subject with facial expression. Numerical validations of the proposed method are given, and experimental results show that the proposed method improves the recognition rate significantly.