基于规范子块五点和的快速分形图像编码

分形图像编码通常需要较长的时间,编码时间主要花费于在一个海量码本中搜索每个输入子块的最佳匹配块.针对这个问题,提出一个限制搜索空间的算法.它主要基于图像块的一种新特征——五点和,把搜索范围限制在初始匹配块(五点和意义下与输入R块最接近的D块)的邻域内.实验表明:该算法能够大大减少子块匹配比较的数量,与基于叉迹的快速分形算法比较,在相同的搜索邻域内,在编码时间、图像质量和压缩比方面都更优.

Fast Fractal Image Coding Based on Quincunx Sums of Normalized Blocks

Fractal image coding can provide a high decoded image quality at a high compression ratio, but it requires traditionally a very long runtime in the encoding process. Therefore, it is essential to develop fast encoding algorithms before it could be widely used for various applications. The encoding time is mostly spent on searching for the best-matched block to an input range block in a usually-large domain pool; a new scheme is thus proposed to limit the search space in this paper. It first defines a new feature, quincunx sum, which is the intensities sum of a normalised image block over each corner of the block and one at the centre. Then, the quincunx sum is utilized to confine efficiently the search space to the vicinity of the initial-matched block (i.e., the domain block having the closest absolute quincunx sum to that of the input range block being encoded). Experimental results show that this method can reduce drastically the amount of range-domain comparisons needed to encode each range block. The proposed algorithm has also been compared with the fast fractal encoding algorithm based on cross-trace, showing that under the same search neighborhood it performs better in terms of encoding time, image quality and compression rate.