Abstract
Fractal image coding is a novel and attractive technique for still image compression. By utilizing the characteristic of self-similarity, an iterated function system can automatically converts an image into a set of affine transformation coefficients. Under the constraint of contractivity, a decoded image that is similar to the original one can be reconstructed by applying this iterated function system onto any arbitrary initial image. This property causes fractal image coding to have the potential of encoding an image with extremely low bit rate if the number of affine transformation coefficients is carefully controlled and encoded. To reach such goal, a scheme of flexible partition of image is essential first. In the past years, several low bit rate methods were proposed. However, even some of the results employ quadtree segmentation technique to improve the compression ratio, the performance of them still cannot satisfy the requirement of very low bit-rate. In this dissertation, we propose a thorough fractal image compression system to approach the target of very low bit-rate. By observing the behavior of other fractal image coding methods, we realize that the block-wise segmentation of image inadequately satisfies the property of natural images. Therefore, an image dependent region-based segmentation technique that attempts to achieve a better performance is proposed. This region-based procedure consists of two steps: First, we improve the performance of quadtree decomposition by utilizing the adaptive threshold method. Second, a merging scheme is introduced to the result of quadtree decomposition that combines several similar blocks into a small number of regions. We also provide a quadtree-based segmented chain code to efficiently record the contours of the regions. Moreover, a post-processing algorithm is applied according to region-based segmentation to eliminate the blocking artifact. In the coding process, we consider both the bit rate and the image quality (in PSNR) simultaneously to reconstruct the image. The experimental results show that the proposed method has the potential to compete with other compression methods to achieve the lowest bit rate at the same level of quality.