基于稀疏分解的可伸缩图像编码

Scalable Image Coding Based on Sparse Decomposition

图像编码技术的新的突破可寄希望于信号表示的深刻变革.采用基于冗余原子库的快速匹配追踪算法对图像进行稀疏分解,在分析和总结原子空间位置分布规律的基础上,提出与之相适应的块划分编码方法,节约了用于表示原子参数和投影系数的比特数.实验结果表明,本文编码器在计算复杂度、编码效率和伸缩性能等方面都优于当前同类型编码器,特别是在前两方面,其优势十分明显.比如对512×512测试图像,编码率为0.5bpp时本文编码器的平均PSNR增益达1.73dB.特别地,凭借原子库的几何特性,该编码器提供了较传统方法更灵活的伸缩性,允许通过简单的参数变换来获得任意分辨率大小的重建图像.

New breakthroughs in image coding may rely on deep changes in the signal representation.The fast matching pursuit（MP） algorithm is first employed to get the sparse image decomposition over a redundant dictionary.The distribution of the spatial position of selected atoms is studied and a novel block partitioning coding method is then proposed.The bit savings in representations of atom parameters and projection coefficients are obtained.Experimental results show that the new coding scheme has striking advantages over the latest MP coder in computational complexity,coding efficiency as well as scalability.For instance,for 512×512 test images an average PSNR gain of 1.73dB is achieved at 0.5bpp.Notably,thanks to the geometrical structure of the dictionary,the new coder provides attractive adaptability features which allow the codestream to be easily and efficiently decoded at any spatial resolution.