Search

As an efficient tool for image compression, wavelet has been widely used in all kinds of image processing areas. Based on the different encoding effects, wavelet compression algorithms can be probably classified into two categories. They are the embedded wavelet coding algorithms and the non-embedded wavelet coding algorithms. For the convenience of producing the anytime cut coding stream and the progressing reconstruction results, the embedded wavelet coding algorithms have been paid more attention in practice. Such as the embedded wavelet coding algorithms, EZW and SPIHT are the outstanding representatives. The only drawback for this wavelet based embedded coding algorithms is the choice of the different wavelet transform base. We propose a novel embedded coding algorithm based on the reconstructed DCT coefficient to avoid the difficulties brought by the choice of wavelet transform base in this paper. The new algorithm΄s efficiency can be seen from the experimental results. [ABSTRACT FROM AUTHOR]

In this paper, a double domain based robust digital image watermarking scheme is proposed. Up to now, the existing robust watermarking algorithms can only resist a part of geometrical attacks and some common image processing attacks, while the proposed method can resist almost all geometrical attacks and a lot of common image processing attacks. Firstly, the watermark is embedded in the 3-level DWT coefficients by the parity modulation method, and then the same watermark is embedded in the pixel domain by the similar method. At the detector side, the watermark is extracted twice, one is in the DWT domain, the other is in the pixel domain. Select the watermark which has better visual quality as the final result. By carefully regulating the parameters and selecting the embedding sequence, the watermarks in the two domains are compatible with each other. Experimental results show the effectiveness of the proposed method. [ABSTRACT FROM AUTHOR]

This paper proposes a scheme to hide two watermarks in visual secret sharing of a gray level image. One watermark is embedded during secret image halftoning, and the other is hidden during the halftone image encryption. A non-expansion visual secret sharing model is used to encrypt the halftone image. In watermark extraction, only Exclusive-OR operation is needed, and basic properties of conventional visual cryptography technique are still preserved. Experimental results show the effectiveness of our schemes. [ABSTRACT FROM AUTHOR]

In the digitized world nowadays, digital content (audio, images, video, 3D meshes, etc.) can be easily copied, manipulated and distributed. Copyright protection and integrity verification of digital content have become urgent problems for multimedia owners and distributors. In addition, to alleviate bandwidth requirements, vector quantization (VQ) is an effective and popular vertex data compression technique for triangle meshes. In this paper, we present a new data hiding method for 3D triangle meshes in the predictive VQ domain. The proposed method is reversible and enables the cover mesh data to be completely restored when the payload is removed from the VQ bitstream. The mechanism is embedding the payload by modifying the prediction rules during the VQ process. Besides, the hash of the cover mesh can be hidden for the self authentication purpose. Experimental results demonstrate the high capacity of the proposed data hiding scheme. [ABSTRACT FROM AUTHOR]

Texture segmentation is an important component in texture analysis. Gabor wavelet shows potential capacity in describing the texture. Kernel-based methods have demonstrated excellent performances in a variety of pattern recognition problems. This paper extracts texture features in the Gabor wavelet transform domain and segments the computed feature image with kernel-based techniques, Spectral Clustering Algorithm (SCA) and Support Vector Machine (SVM). Due to the severe computational complexity of kernel-based techniques, we split the main segmentation process into two steps. SCA is firstly applied to the sampled feature image. Then the clustering result serves as training samples and is used to train SVM. The initial segmentation result is obtained by labeling the feature image with the trained SVM. A median filter is employed to improve the segmentation result. Four mosaic texture images and a natural scene are used to test our new algorithm. [ABSTRACT FROM AUTHOR]