在综合分析当前图像压缩算法的基础上，提出新的基于分层变块大小分解的图像压缩构想。JPEG、JPEG2000、分形作为当前最为流行的3种静态图像压缩算法，在对不同的图像进行相同倍率的压缩时，表现出同样的性能趋势：视觉上越复杂的图像，恢复图像的质量越低。经过大量实验发现，3种算法的压缩性能均与同一个指标存在明确关系——图像活跃度量(IAM)。根据图像不同区域的复杂程度不同，采用IAM和相似度作为性能指标，利用粒子群优化(PSO)算法求解最优近似图像，实现对图像的分层变块大小分解（SVBD），将图像中相同复杂特性的区块归为一类。该分解方式符合人类认知图像内容的特点，为提高压缩性能创造了有利条件。

A concept of image compression based stratified variable blocksize decomposition is proposed after analyzing current image compression algorithms compreehensively. As the current most popular methods for static image compression, JPEG, JPEG2000 and fractal perform similarly when compressing different images with the same compression ratio: the more visually complex an image is, the lower a restored image quality is. After a large number of experiments, previous work shows that there is a clear relationship between all the three methods and image activity measure (IAM). According to the complexity of different image regions, stratified variable blocksize decomposition (SVBD) is achieved, using IAM and similarity as performance index. Particle swarm optimization (PSO) algorithm is used to find optimal approximation of the image blocks. Finally, by classifying the image blocks, some improvement to compression quality can be made.