遥感立体、多视图影像压缩研究

Stereo image pairs, three-line-scanner and other multi-view image are widely used in mapping and remote sensing. Compared with monocular imaging system, stereo and multi-view imaging system produce several times more data than monocular system with same resolution. However, there exists high correlation between each two views. The bit-rate of pixels by exploiting efficiently the correlation existed in these views can be lowered to alleviate the pressures of the storage system and transmission system. However, the stereo and multi-view image compression technology is still on the primary phase, far from being mature as the common image compression technology. There is still some distance away from the actual application for multi-view image compression technology. This project will focus on the compression of stereo and other multi-view image by incorporating wavelet transform with adaptive technology and predictive coding theory. The main work includes the following aspects..i) Vector lifting schemes (VLS) can exploit the correlation existed in multiple views. But the performance was seriously impaired in occlusion area. Considering the good performance and existing problem of VLS, an adaptive lifting scheme is introduced for stereo image compression by incorporating VLS with spatial wavelet transform. The VLS incorporate disparity compensation with the decomposition of target images, in other words, the information of reference image is directly applied in the decomposition of the target image. However, due to the occlusion, the reference image can not always provide a good prediction for the coefficients of sub-bands of target image. As a result, the decorrelation performance of the decomposition of VLS will be greatly affected. Instead, the intra-view wavelet transform is exploited to avoid the effect of occlusion. In addition to stereo image pairs, the adaptive lifting scheme is extended to the compression of three-line-scanner image and other multi-view image compression..ii) After a comprehensive analysis of the existing lossless image compression algorithm, stereo and multi-view image lossless compression algorithms will be covered. At present, lossless image compression algorithms consist of predictive coding methods and reversible wavelet transform-based lossless compression methods. Considering the intra-correlation and inter-correlation of multi-view image, a lossless predictive coding method is proposed for stereo and multi-view images.

立体像对、三线阵及多视影像被广泛应用于测绘、遥感领域。和单幅成像系统相比，多视成像系统的数据量在分辨率相同的情况下要多出许多倍。然而多个视图之间存在着较高的相关性，高效利用多视影像之间的相关性可以进一步降低像素的比特率，从而减少传输和存储的压力。本课题以立体像对及其他多视图影像为研究对象，综合运用小波变换、自适应技术及预测编码理论研究立体、多视图影像的压缩技术。矢量提升模式能够充分利用多视之间的相关性，然而遮蔽区域(occlusion)根本不存在相关性。本课题采用一种自适应的提升分解方法，根据局部区域特性自适应地选择矢量提升分解还是空域小波分解，从而获得更大的编码增益。为了更全面地优化压缩算法，研究建立能够反映密集匹配优劣的综合压缩评价方法与理论。此外，还将研究综合像内与像间预测的立体及多视图影像无损预测编码方法。

近年来，对地观测系统被广泛应用于国防、农业生产及减灾防灾等领域。要建立高水平的对地观测系统，高质量的影像数据获取是首当其冲的问题。特别是星地间不稳定的数据传输链路和越来越高的影像分辨率使得影像的及时传输成为一个挑战。因此，针对各种类型的影像数据的压缩编码成为当前的一个研究热点。本课题以测绘、遥感领域必不可少的立体像对及三线阵等多视图影像为研究对象，综合运用小波变换、自适应技术及预测编码理论研究立体像对及三线阵等多视图影像的压缩编码方法。.在有损压缩方面，设计了一种基于小波变换和率失真优化的自适应视差补偿的立体图像和三线阵图像的压缩方法。高效利用多个视图之间的相关性是多视图影像压缩的关键。然而，在遮蔽区域或者纹理极为复杂的区域，多个视图之间往往不存在相关性，因此设计了基于率失真优化的视差补偿来根据区域特性自适应的消除视图相关性。实验结果表明，从压缩性能上来看，本方法与基于H.264（视频压缩标准）的方法相当，尽管低于基于H.265（最新视频压缩标准）的方法，但是从时间复杂度上来看，本方法的时间复杂度是H.264的一半，仅为H.265的1/20。因此，该方法是一个更为实用的立体及三线阵图像压缩方法。.此外，为了将常用的预测编码方法用于遥感影像的近无损压缩，设计了基于分块容差调整的码率控制算法。为了获得可伸缩的压缩码流，还设计了一种基于小波域逐层自适应视差补偿的立体图像编码方法。实验结果表明，与采用JPEG2000独立编码比较，本算法获得了显著的编码增益。.在无损压缩方面，提出了综合利用视图间与视图内相关性的立体图像无损预测编码方法。与现有的无损图像比较，本方法获得了显著的编码增益。此外，对支持分辨率渐进的无损图像编码方法也作了研究，提出了一种新颖的图像分解去相关方法，结合小波提升模式与边缘自适应预测研究实现了一种比二维小波变换性能更好的分解方法。与JPEG2000无损编码模式比较，本算法获得了显著的编码增益。.本课题还研究设计了一种基于小波域梯度相似性的图像质量评价方法。实验结果表明，与现有模型对比，该评价指标更好地体现了主观与客观评价的一致性。对图像压缩方法的设计起到了一定的指导作用。