Catálogo de publicaciones - libros

The paper presents a reversible watermark scheme based on hash function and lossless compression technology. The paper extracts the abstract information of the image as hash code and lossless compresses the LSB(Least Significant Bit) of the image to make some space where the image hash can be completely inserted. It needs to be done without introducing a large distortion to the image. The scheme can achieve the authentication purpose by hash code and restores the original image by retrieving the watermark sequence completely because of the lossless decompression of the LSB of the image.

In recent years, the method of spread spectrum watermarking () has become the one of the most promising technique used for digital watermarking technology. However, the major problems of SSW technology is that the effectiveness of detection will drop fairly without subtracting the host image from watermarked image because of the effect of correlation remains calculated between the host image and original pseudo-random sequences. In our previous work, we have proposed a method to improve the detection of SSW techniques by reduction of correlation remains using pixel-correlations in an image. In this paper, we have investigated the effectiveness of the proposed method with image experiments in detail. The experimental results illustrate that the correlation remains can be reduced greatly by using pixel-correlations.

This paper proposes an effective technique which can detect malicious manipulations under video lossy compressing data (e.g. H.263) and still-image lossy compressing data (e.g. JPEG). A block-classification strategy is used to divide DCT-blocks into the flat-blocks and the normal-blocks. Simple features of the both blocks are embedded invisibly. For later authentication, the watermarked frame is then put back into the compressed bit-streams.

The proposed method can detect and locate alterations of the tampered frame whether it is stored again into original bit-streams or not. This goal is for detecting tamper area and surviving most video lossy compression. Experimental results show that the proposed technique can detect various tampered areas and hence provides an effective image authentication for lossy compressed image and video in DVR (Digital Video Recorder) System.

Semantics-based image retrieval is a challenging problem. In this paper, we propose an approach for image semantics abstraction, which constructs a multi-level semantics tree based on human subject and train hierarchical semantic classifier. According to our method, image features are selected by using priori knowledge. Then, those images are classified in every level by the classifier based on support vector machines (SVM). The SVM classifiers learn the semantics of specified classes from a training database of image. Experiments show that we can abstract multi-level semantics from image database by using less low-level features.

Scene detection is an essential step to organize the video data properly for content-based video analysis and its application. In this paper, an effective scene detection approach is proposed, which exploits the cinematic rules used by filmmakers as guideline to compute shot similarities and identify the video scenes in narrative film. First, a clustering method with time constraint is used to group the shots into scene slices. Then dialog scene’s alternative structure and audio correlation are used to identify dialog scene. Finally, motion and audio correlation guided by cinematic rules are used to further detect action scene. Experimental results show that the proposed method works well and can deal with complex scene.

This paper presents a motion suppression technique for video shot boundary detection based on 3D wavelet transform (3-DWT). Dramatic motion can be characterized in terms of energy and variance, and be differentiated from various video effects. Motion suppression value (MSV), which indicates intensity of motion energy, is extracted and used to suppress motion influence suffered by traditional video shot boundary detection algorithms. Unlike previous methods, our technique is robust to dramatic motion inherent in video sequence, experimental result validates our approach.

In this paper, we propose a Generalized Probabilistic decision based Neural Network (GPDNN) for content-based image retrieval (CBIR). Instead of receiving the numerical values of each data points as the input, the proposed GPDNN models the I/O relationship via the distribution of input data and their corresponding outputs. The GPDNN involves the Multiple-Instance learning techniques to learn a desired concept. A set of exemplar images are selected by a user, each of which is labeled as conceptual related (positive) or conceptual unrelated (negative) image. Then, by using the proposed learning algorithm, an image classification system can learn the user’s preferred image class from the positive and negative examples. The experimental results show that for only a few times of relearning, a user can use the prototype system to retrieve favor images from the database.

In this paper, we present a new method for classifying shot type in sports video using fuzzy information granular. The problem is important for applications such as video structure analysis and content understanding. In particular,two-stage off-line learning processes perform knowledge extraction of semantic concepts and automatic shot classification, respectively. In the first stage, the extracted prominent regions are used as a good pattern in semantic concept level. Then a number of global features are defined as efficient input of the shot type classifier in the second stage. The identification of semantic concepts and classification of shot are based on soft decisions. Hence, this framework can adequately capture the uncertainty or ambiguity of scales of a shot. Experimental results show the excellent performance of the approach.

Searching for similar images is an important research topic for multimedia database management. This paper uses a quality index model to search for similar images from digital image databases. In order to speed up retrieval, the quality index model is partitioned into three factors: loss of correlation, luminance distortion, and contrast distortion. The method is performed on three different image databases to test for retrieval accuracy and category retrieval ability. The experimental results show that the proposed method performs better than the color histogram method, the color moment method, and the CDESSO method.

Image registration is a crucial step in all image analysis tasks. In this paper, an edge projection-based image registration method is proposed. First, Radon transform is used to project the edges of images along different directions, then cross-correlation-based approach is used to estimate global rotation and shift, finally polynomial fit is used to improve the accuracy of image registration. Simulation indicates that the proposed method can estimate global rotation and shift accurately and speedily.