A Fast Fractal Encoding Method Based on Fractal Dimension

Abstract

In this paper a fast fractal coding method based on fractal dimension is proposed. Image texture is an important content in image analysis and processing which can be used to describe the extent of irregular surface. The fractal dimension in fractal theory can be used to describe the image texture, and it is the same with the human visual system. The higher the fractal dimension, the rougher the surface of the corresponding graph, and vice versa. Therefore in this paper a fast fractal encoding method based on fractal dimension is proposed. During the encoding process, using the fractal dimension of the image, all blocks of the given image first are defined into three classes. Then each range block searches the best match in the corresponding class. The method is based on differential box counting which is chosen specifically for texture analysis. Since the searching space is reduced and the classification operation is simple and computationally efficient, the encoding speed is improved and the quality of the decoded image is preserved. Experiments show that compared with the full search method, the proposed method greatly reduced the encoding time, obtained a rather good retrieved image, and achieved the stable speedup ratio.

Impact of Spatial Dynamic Search With Matching Threshold Strategy on Fractal Image Compression Algorithm Performance: Study

“…In another work, Wang, et al [35] have used the standard deviation feature to segment the range blocks into two classes, where the range blocks of only one class are encoded by searching of the suitable matched domain block with the help of particle swarm optimisation technique, while the range blocks in the other class are encoded directly without search by storing their average values. Similarly, but with more defined classes, the research articles [10], [22], [36], [37] have introduced different methods to divide the image blocks into groups, and restricting the search within the blocks of the same class only. Quite similar to this approach is the approach of neighbourhood search, or nearest neighbour search, where the search is restricted to a small set of the domain blocks that share similar feature vector with the corresponding range block [20], [21], [38], [39].…”