基于多原子快速匹配追踪的图像编码算法

Multi-atoms Rapid Matching Pursuit-Based Image Coding Algorithm

该文提出一种多原子快速匹配追踪信号稀疏分解算法,并将其应用于静态图像编码。多原子匹配追踪通过每次迭代选取多个原子的形式,实现信号的快速稀疏分解。在此基础上,通过构造多尺度脊波字典实现图像的稀疏分解,并对稀疏分解的数据进行自适应量化和编码。实验结果表明,多原子匹配追踪获得了与匹配追踪相当的逼近性能,同时极大地提高了稀疏分解的速度。新的编码算法在低比特率情况下,获得了比JPEG2000更理想的编码性能。

In this paper, a Multi-Atoms rapid Matching Pursuit （MAMP） algorithm for signal sparse decomposition and its application to image coding are proposed. The MAMP algorithm decompose signal sparsely by selection several atoms at each iteration. A multiscale ridgelet dictionary is constructed and used to represent image based MAMP. The sparsely decomposed data are adaptively quantized and encoded. Experimental results show that the approximation performances of the MAMP algorithm are comparable with those of the matching pursuit. Meanwhile, the computation speed is greatly improved. On the other hand, the performances of the new coding scheme are shown to compare favorably against those of the state of the art JPEG2000 scheme at low bit rate.