面向遥感图像高保真压缩的变换与量化方法研究

With the continuous improvement of remote sensing image resolution, the conflict between the dramatically increasing image data and the limited transmission bandwidth and storage space is becoming more and more prominent, and has a serious impact on the timeliness and efficiency of the remote sensing images application. Image compression technology as an effective way to solve the real-time transmission and storage of massive remote sensing image data, faces enormous opportunities and challenges. For the low bit-rate and high-fidelity compression of remote sensing image, the greatest difficulties lie in the complex image texture, the weak and small targets, and the requirement of high speed real-time processing. The low bit-rate compression of remote sensing images with the traditional compression methods can not meet the requirements of the quality of reconstructed image. This project aims to research the transform and quantization methods for low bit-rate and high-fidelity compression of remote sensing images, and designs efficient hardware structure of the compression algorithm. intended to carry out research work in the following ways: Firstly, based on the concentration of energy gain and the quality of reconstructed image, a wavelet basis optimization criterion is proposed, and the optimal bi-orthogonal wavelet bases for remote sensing image is constructed, to achieve high-fidelity compression transform. Secondly, in wavelet domain, a quantization model based on the sensitivity of visual distortion is created, and the adaptive quantization method is designed to improve the overall subjective quality of reconstructed images. Finally, an efficient hardware structure for the transform and quantization algorithm is designed to solve the high-speed real-time compression of remote sensing images.

随着遥感图像分辨率的不断提高，急剧增加的图像数据与有限的传输带宽、存储空间之间的矛盾越来越突出，已严重影响了遥感图像应用的时效性。遥感图像压缩技术作为解决海量图像数据实时传输与存储的有效途径，面临着巨大的机遇和挑战。由于遥感图像纹理复杂、目标相关性弱，压缩处理的实时性要求高，使用传统压缩方法进行高倍压缩时，在图像重建质量、目标保持能力上难以满足实际应用的要求。本项目旨在研究遥感图像高倍高保真压缩中的变换与量化方法，以及压缩算法的高效硬件设计与优化技术，拟在以下三方面开展研究工作：1）提出以能量集中增益和图像重建质量为目标的小波基综合优化准则，构造基于遥感图像特征的最优双正交小波基，实现高保真压缩变换；2）提出基于小波域视觉失真敏感性的模型，设计小波域的自适应量化方法，提高恢复图像的整体主观质量；3）研究变换及量化算法的高效硬件处理结构，解决遥感图像高保真压缩的高速实时处理问题。

可见光遥感图像是高分辨率对地观测的主要载体，在国防、工农业、经济和科研等众多领域发挥着重要的作用。随着成像能力的提高，海量遥感图像与有限传输带宽之间的矛盾已逐渐成为制约遥感技术发展的瓶颈，遥感图像压缩是解决这一矛盾的有效手段。由于遥感图像具有不同于普通景物图像的应用场景、成像特点和数据特征，因此，在研究相应压缩技术的过程中应设计有针对性的算法。本项目旨在研究遥感图像高倍高保真压缩中的变换与量化方法，以及压缩算法的高效硬件设计与优化技术，在以下几方面开展了相关研究工作：设计了以能量集中增益和图像重建质量为目标的小波基综合优化准则，构造基于遥感图像特征的最优双正交小波基，实现高保真压缩变换；设计了基于小波域视觉失真敏感性的模型，设计小波域的自适应量化方法，提高恢复图像的整体主观质量；研究了变换及量化算法的高效硬件处理结构，对遥感图像高保真压缩的高速实时处理问题提出了较好的解决方案。通过该基金项目的支持，项目组在二维方向、自适应量化和算法的VLSI综合优化等方面取得了一系列研究成果，发表了 SCI/EI 论文共计15篇，申请专利6项，培养了9名研究生。