稀疏度拟合的自适应图像并行压缩感知算法

Adaptive Image Parallel Compressed Sensing Algorithm Based on Sparsity Fitting

为了提高图像的重构精度和处理速度,提出一种稀疏度拟合的自适应小波包图像并行压缩感知算法.首先采用小波包对大小相同但不重叠的图像块进行稀疏变换,在最优分解尺度下利用迭代方法确定满足图像重构精度的最低采样率,并采用最小二乘法对采样率进行优化处理;然后结合云计算技术,利用MapReduce框架对算法进行并行化.在实验室构建Java开发环境下的计算机集群,采用标准图像作为样本比较不同算法的压缩率、重构性能和运算时间,结果表明,该算法的重构质量和处理速度均得到显著提升.

In order to improve the reconstruction accuracy and processing speed of an image,an adaptive wavelet packet image parallel compressed sensing algorithm with sparsity fitting is proposed.First,the sparse transformation was carried out on the image blocks which were of the same size and not overlapped using wavelet packet.An iterative method was used to determine the minimum sampling rate satisfying the accuracy of image reconstruction under the optimal decomposition scale,and the least square method was used to optimize the sampling rate.Then,the algorithm was parallelized with MapReduce framework combined with cloud computing technology.A computer cluster was built in the laboratory under Java development environment,and the compression rate,reconstruction performance and operation time of different algorithms were compared by using standard image as samples.The results show that the reconstruction quality and processing speed of the algorithm are improved significantly.