Department of Electronic Engineering, Hong Kong Polytechnic, Kowloon, Hong Kong.
IEEE Trans Image Process. 1998;7(6):888-91. doi: 10.1109/83.679437.
In fractal image block coding, most of the time is spent on finding a close match between a range block and a large pool of domain blocks. For a large image, this effect becomes aggravated as the domain pool increases exponentially. We propose using the local variances of domain blocks to reduce the search space. By sorting the contracted domain pool according to their local variances and defining an acceptance criterion for a close match, we can confine all the potential close matches to a relatively small sized window to limit the search space. The encoding time can hence be shortened with the decoded image quality as good as that using the full search method. The speedup can be over ten times depending on the complexity of encoded images.
在分形图像块编码中，大部分时间都花在查找范围块与大型域块池之间的紧密匹配上。对于大图像，随着域池呈指数级增长，这种效果会变得更加严重。我们建议使用域块的局部方差来缩小搜索空间。通过根据局部方差对收缩的域池进行排序，并为紧密匹配定义一个接受标准，我们可以将所有潜在的紧密匹配限制在一个相对较小的窗口中，以限制搜索空间。因此，编码时间可以缩短，而解码图像质量与使用全搜索方法一样好。根据编码图像的复杂度，加速可以超过十倍。
