Research on Region Adaptive Image Compression Algorithms and Integrated Schemes

研究中获得的主要结果如下：(a)研究了一种新的区域自适应分形图像编码方法，该方法将原始块合并为可变大小范围的块。我们得到了以下结果：在原始块合并过程中，利用块合并引起的整体误差增加，SN比提高了1dB.ii。对于(1)方案，增加的阈值不是恒定的，而是先小数值，然后逐渐增加。当获得按顺序合并的范围块时，以相反顺序与周围块重新合并的方案减少了合并顺序的影响。上述(i) - (iii)方案可使压缩性能提高2dB.v。可选择译码迭代次数的分形图像编码方案表明译码在一定时间内完成。(b)对区域自适应图像压缩的核心——块匹配进行了研究，得到如下结果：提出了一种减少块间MSE计算次数的数学算法。它的计算速度大约是传统方案的3倍。此外，修改搜索顺序和限制搜索范围可以减少一半的处理时间。(c)在依赖于已编码区域的图像压缩方案中，提出了一种新的方法，如果由于编码数据延迟或迭代中断而导致解码过程未完成，则将数据完全解码，并修改内部缓冲区，直到开始下一个区域的解码过程。这种方法大大减少了误差的传播。

Main results obtained in the research are listed below.(a) We studied new region adaptive fractal image coding in which primitive blocks are merged into a variable size range block. We have obtained following results.i. In the merging process of primitive blocks, the use of increase of overall error caused by block merging results in the improvement of SN ratio by 1dB.ii. It is effective to the scheme of (i) that the threshold of increase is not constant but firstly small values and increased gradually.iii. The scheme where a range block is re-merged with surrounding blocks in reverse order when a range block merged in an order is obtained reduces influence of merging order.iv. The above schemes of (i) - (iii) achieve the improvement of compression performance by 2dB.v. A new fractal image encoding scheme which can select decoding iteration times shows decoding are finished within a constant time.(b) We studied block matching which is the core of the new region adaptive image compression and obtain following results.i. A mathematical algorithm that reduces calculation times of MSE between two blocks is proposed. It calculates about 3 times faster than the conventional scheme.ii. Moreover, modification of searching order and restriction of searching range result in the half of processing time.(c) In image compression schemes which depend on the region already encoded, a new method is proposed that the data is completely decoded and that the inner buffer is modified until decoding process of the next region starts if decoding process is not completed because of a delay of encoded data or a break of iteration. This method greatly reduces error propagation.