基于广义建模理论的多原子库图像编码方法研究

现有的图像编码框架中，以DCT、小波变换为代表的基于基展开的编码方式没有考虑图像信号本身的特征，而以稀疏分解为代表的基于统计建模的图像编码又不能满足模型空间普适性的要求。针对这一问题，本项目通过引入图像广义建模理论进行图像的统计特性分析，并由此开展多原子库模型下的图像编码框架的研究。首先，提出基于图像广义建模理论的信息度量方法，在综合考虑多模型空间带来的图像在选择最优原子库时产生的不确定性以及多原子库带来的图像总体编码增益的情况下，将编码转化为最小化能量函数的优化问题；其次在定义广义建模性冗余的概念后，将图像建立有针对性的原子库的过程等价为图像广义建模性冗余去除的过程，由此提出一种基于特征区域提取的模型空间优化方法；最后，配合相应的参数提取方法，量化策略，熵编码方案以及优化理论，以最小化广义建模能量函数为目标提出一种多原子库图像优化编码框架，从而为图像/视频处理技术的发展开辟一条新的途径。

In the existed image compression frameworks, the basis expansion( i.e. DCT and Wavelets) intends to express any arbitrary signal with a finite function or vector, while neglecting the feature of signal per se, while the statistical modeling(i.e. Sparse coding) of image representation attempts to explore model space which correspondents to the signal feature through analyzing human visual system.In this project, First of all,we introduce the concept of image modelibility and computing method in order to settle the theoretical basis of our research. Image modelibility reveals the image's adaptation capacity to the model space. The adaptation capacity per se, is measured by two factors: the uncertainty generated when images select the most optimized dictionary and The coding gain brought by multi dictionary library.Through introducing the redundancy with modelibility, the conclusion is that the process of extraction of image feature and establishment of targeted dictionary library is of the same equivalence of the eradication of image redundancy. Then we settle the most optimized dictionary library of seven common scenarios or targets', including indoor scenarios, outdoor scenarios, human bodies, human faces, vehicles, plants and constructions. This research is based on image extraction encoding frame of image modelibility theory, i.e. an optimized encoding method, which used minimized energy function based on image modelibility equation, along with the corresponding atom parameters computation method， quantitative strategy, parameter entropy coding and optimized theory.