基于递归预测的分布式双视角图像无损压缩

Distributed Dual-view Image Lossless Compression Based on Recursive Prediction

针对星载双视角图像压缩中存在数据量大和卫星编码端计算能力受限、内存资源不足的问题,提出了一种基于递归预测的分布式双视角图像无损压缩方法。该方法中2个视角的图像采用不同的编码方法,视角1作为关键视角采用JPEG2000无损模式进行独立压缩;视角2经图像分块预处理之后,其中一个局部图像块作为关键图像块仍用JPEG2000无损模式独立压缩,其余部分采用分布式编码方法。编码端进行二维整数小波变换去除各视角内的空间冗余,解码端利用利用视角间的相关性采用递归预测的结构,通过配准和多元线性回归的方法生成视角2其余图像的边信息辅助其余图像块解码。实验结果表明,与未考虑双视角图像间冗余的JPEG2000无损压缩编码相比,平均编码比特率大约节约了0.296~0.6 bpp,时间复杂度降低到其5.97%~14.3%;与使用初始边信息的分布式编码方案相比时间复杂度相同,但平均编码比特率大约节约了0.45~0.51 bpp;与考虑了图像间冗余的相关文献图像算法相比,所提方法的编码比特率损失0.2 bpp,但编码复杂度降低到其4.3%,说明所提方法更具优势,满足星载图像压缩需求。

To deal with the problems of large amount of data,limited computing power and insufficient memory resources at the satellite encoding end in the spaceborne dual-view image compression,a distributed dual-view image lossless compression method based on recursive prediction is proposed.In this method,the images of the two views will use different encoding methods.For View 1 which is used as the key view,independent compression is adopted using the JPEG2000 lossless mode;images of View 2 is preprocessed and divided into blocks,then one of the partial image blocks used as the key image block is still compressed independently using JPEG2000 lossless mode,while the rest uses a distributed encoding method.Firstly,the encoding end performs two-dimensional integer wavelet transform to remove the spatial redundancy in each view,and then the decoding end utilizes the correlation between views and adopts the recursive prediction structure to generate the side information of the remaining images of view 2 through registration and multiple linear regression methods.The side information assists the decoding of the remaining image blocks.The experimental results show that compared with the JPEG2000 lossless compression coding that does not consider the redundancy between the images of two views,the average coding bit rate is about 0.296~0.6 bpp saved,and the time complexity is reduced to 5.97%~14.3%;compared with the distributed coding scheme that uses the initial side information,the time complexity is the same,but the average coding bit rate is saved by about 0.45~0.51 bpp;compared with the related image algorithms in literatures considering the redundancy between images,the coding bit rate loss of the proposed method is 0.2 bpp,however,the encoding complexity is reduced to 4.3%,which shows that the proposed method has more advantages and meets the needs of spaceborne image compression.