Catálogo de publicaciones - libros

Video shots provide the most basic meaningful segments for video analysis and understanding. In this paper, we present a detection and classification framework for the video shot segmentation in a coarse-to-fine fashion. The initial transitions are detected from a sub-sampled video space. These coarse segments are later refined in the original video space with the technique of illumination artifacts removal and transition finalization. The transition type (abrupt or gradual) are finally determined by exploiting the histogram intersection plot. The proposed method has been tested on a large amount of videos, which contain a variety of types of shot transitions. Accurate and competitive results have been obtained.

In all-in-focus imaging, a series of photographs taken of the same objects, on different focal planes, are analyzed to create an entirely in-focus final image. Edge detection techniques determined by variable thresholds are applied to the 512x512 input images and they are then progressively subdivided into smaller, 2 sized blocks, varying in size from 256x256 pixels down to 1x1 pixel. The 1x1 blocks are used to determine actual edges and the areas around them are then filled with progressively larger block sizes. The particular image that is most in-focus over a given region is determined by comparing the sums of edge pixels for the corresponding blocks of the different input images. Beyond aesthetic value, all-in-focus imaging may be used in applications ranging from pattern recognition and object detection to biometrics. We have successfully detected in-focus regions in an image and have generated final, all-in-focus images with only minor errors.

Form document image processing has become an increasingly essential technology in office automation tasks. One of the problems is that the document image may appear skewed for many reasons. Therefore, the skew estimation plays an important role in any automatic document analysis system. In the past few years, many algorithms have been developed to detect the skew angle of text document images. However, these algorithms suffer from two major deficiencies. Firstly, most of them suppose that the original image is monochrome and therefore they are not suitable to apply to documents with a complicated background. Secondly, most of the current methods were developed for general document images that are not as complicated as form documents. In this paper, we present a new approach to skew detection for grey-level form document images. In our system, image decomposition by 2D wavelet transformations is used to estimate the skew angle.

A neglected challenge in existing e-Learning (eL) systems is providing access to multimedia to all users regardless of environmental conditions such as diverse device capabilities, the heterogeneity of the underlying IP network, and user modality preference. This paper proposes a novel two-tier transcoding framework capable of adapting eL multimedia to meet the end-user environmental challenges. This two-tier architecture consists of 1)an application layer transcoder that adapts the presentation format of the eL content as viewed in a browser to meet device capabilities and user modality preference, 2) a bitstream transcoder that transforms multimedia streams to conform to the device’s processing capabilities and to adapt the encoding rate to meet the network’s fluctuating bandwidth. Results demonstrate the eL multimedia transcoding for mobile devices and its low overhead delays.

The presence of highlight can lead to erroneous results in Computer Vision applications such as edge detection, and motion tracking. Many algorithms have been developed to detect and remove highlight. In this paper, we propose a simple and effective method for detecting and removal of highlight. We first use a window to help to remove the noise and reduce the data amount for analysis. We then apply K-means algorithm in a 5-D vector space to computer diffuse chromaticity. In the case of non-white illuminant, illuminant chromaticity is estimated in the inverse-intensity space, and we use Fuzzy C-mean clustering and linear fitting to get illuminant chromaticity. Finally, we use Specular-to-Diffuse mechanism to separate specular reflection component from image. Experiments show that it is robust and can give good results.

As digitalized content rapidly proliferates in networked systems, content security necessarily arises as one of the most important issues. Many developers have studied techniques for allowing only authorized persons to access content. Recently, video scrambling techniques, one type of the authorizing tools, have been introduced. However, they change the original video data, which often increases the bit rate of the source data. To overcome this problem, we propose a scrambling technique which deliberately distorts the original video sequences in a reversible way by arbitrarily relocating the differential motion vectors and MB (macroblock) starting positions in a slice. This method can be applied to most common video coding techniques such as MPEG-1/2/4, and H.264.

The validity of using the weighted information entropy to estimate the complex degree of the infrared images’ backgrounds is discussed in this paper. A Butterworth high pass filter whose cut-off frequency can be adaptively regulated to meet the different backgrounds is proposed to restrain the different clutters. Since the backgrounds usually have some random change in the video sequences, an adaptive binarisation criterion for small target detection is also presented. Experimental results show the robustness of our method.

A new neural network adaptive switching median (NASM) filter is proposed to remove salt-and-pepper impulse noise from corrupted image. The algorithm is developed by combining advantages of the known median-type filters with impulse detection scheme and the neural network was included into impulse detection step to improve its characteristics. Comparison of the given method with traditional filters is provided. A visual example is given to demonstrate the performance of the proposed filter.

With the rapid growing amount of multimedia, content-based infomation retrieval has become more and more important. As a crucial step in content-based news video indexing and retrieval system, shot boundary detection attracts much more research interests in recent years. To partition news video into shots, many metrics were constructed to measure the similarity among video frames based on all the available video features. However, too many features will reduce the efficiency of the shot boundary detection. Therefore, it is necessary to perform feature reduction for every decision of shot boundary. For this purpose, the classification method based on rough sets and fuzzy c-means clustering for feature reduction and rule generation is proposed. According to the particularity of news scenes, shot transition can be divided into three types: cut transition, gradual transition and no transition. The efficacy of the proposed method is extensively tested on more than 2 h of news programs and 98.0% recall with 96.6% precision have been achieved.

Based on the directionality of images and combining the direction information with multiple resolution analysis, an image enhancement idea via fusion based on directional filter banks is presented in this paper. Combining with LP analysis further, an image fusion method based on LPDFB is given in the paper. Using the experiments to compare the results, they prove its feasibility and validity.