基于稀疏表示的人脸图像压缩方法

Facial Image Compression Based on Sparse Representations

基于稀疏表示的人脸图像压缩算法首先对人脸图像进行分块,其次利用K-SVD字典学习算法,训练一个图像的冗余字典,最后用OMP算法对其进行稀疏编码,得到压缩的图像.由于OMP算法复杂度较高,为了降低复杂度,提高算法效率,提出了一种基于稀疏表示理论的新的人脸压缩算法.该算法在稀疏编码阶段,用基于块坐标松弛(Block Coordinate Relation)字典学习算法对人脸图像进行稀疏编码,最后用重构算法对压缩数据进行重构.通过实验仿真,与JPEG压缩方法及OMP算法比较,所提方法在同等压缩比下,重构的图像质量有所提高.

The use of sparse representations in compressed facial images is common in recent years. Firstly, divide the facial image into fixed-size square patches. Secondly, use the K-SVD al- gorithm for training a redundant dictionary, and finally use sparse coding OMP algorithm to obtain a compressed image. The complexity of the OMP algorithm is rather high; in order to reduce it, and improve the efficiency of the algorithm, this paper proposes a new facial compression algorithm based on sparse representation theory. The algorithm uses the Block Coordinate Relaxation diction- ary learning algorithm for facial image sparse coding. And finally , the compressed data is recon- structed by the reconstruction algorithm. The simulation results show that the method has better image quality under similar compression ratio compared to the JPEG and OMP approaches.