Image data compression and image watermarking are two important subareas in multimedia signal processing. This thesis is about on adaptive image compression and adaptive image watermarking algorithms. The main contributions in this thesis are as follows: 1. Two adaptive BTC algorithms are presented. By using edge-based segmentation and contrast features of human vision system (HVS), we improve the compression performance efficiently while remaining the main merits of conventional BTC. 2. We propose a JND-based classifier used in classified block coding that utilizes the visual features of HVS. The classifier is almost as efficient as the most commonly utilized classifier based on the local variance, but has lower computational complexity. 3. Motivated by the analysis on some classifier commonly used, we propose an integrated classifier that is composed of JND-based classifier, the polynomial approximation error based classifier, and the spatial-frequency spectrum based classifier. The integrated classifier performs much better than those single parameter classifiers. The algorithm applying the classifier achieves high compression ratio. 4. By analyzing visual models, we derive a novel quadtree-based segmentation criterion based on luminance masking. The criterion measures the uniformity of a block by using visual sensitivity and so matches the visual features of HVS well. A hybrid coding algorithm applying above criterion is then presented. The algorithm exhibits much better performance than JPEG. 5. We analyze the blocking effects in spatial domain and then propose an adaptive space-variant postfiltering algorithm. The algorithm improves both the subjective and objective quality of reconstructed images efficiently. 6. We find a feature pattern of artifacts in frequency domain and the notch filtering of the edge image extracted from a decoded image. An adaptive postprocessing algorithm based on a neighborhood analysis on the notch- filtered edges is proposed. 7. A classification-based image watermarking scheme in DCT domain is proposed. It is adaptive in that different strength of watermark is applied to different blocks according to block classification. In the classification, luminance and texture masking are taken into account. The invisible watermarks thus generated are robust against common signal processing and noise. 8. Based on a quantitative analysis on magnitude of DCT components, a new embedding strategy for image watermarking is proposed. Different from the popular strategies, we argue that more robustness can be achieved if watermarks are embedded in DC components since DC components have much larger perceptual capacity than any AC components. 9
修改评论