基于TV-泊松奇异积分联合先验模型的图像重构

目的针对当前图像重构算法容易产生过渡平滑图像纹理区域,使复原图像丢失大量纹理,降低重构图像视觉质量等缺陷,提出TV-泊松奇异积分联合先验模型耦合贝叶斯推理的图像重构算法。方法引入配分函数,结合TV函数,构造TV图像先验。定义泊松奇异积分先验,并将其嵌入到TV先验中,设计一种联合先验模型,控制图像纹理平滑度。基于高阶统计量技术,完善图像退化模型,并耦合先验模型,生成重构图像的最大后验估计MAP。引入优化最小原则,求解MAP,完成贝叶斯推理,获取重构图像。对文中算法复原图像纹理的关键参数进行优化,并研究分析该算法的用户响应。结果与当前图像重构算法相比,文中算法的复原视觉质量更高,能够较好地平衡噪声与纹理。在图像退化程度较大时,文中算法具有良好的用户响应。结论文中算法能够较好地同步保持图像边缘与纹理。

Image Reconstruction Algorithm Based on TV-poisson Singular Integral Joint Priori Model

In order to solve these defects such as low visual quality of restoration image induced by over-smoothing textured areas resulting in eliminating image texture in current image restoration algorithms, the image restoration algorithm based on TV-Poisson singular integral joint priori model coupled with Bayesian inference was proposed. TV image prior was constructed by introducing partition function and combining TV function. A new joint priori model was designed by introducing defining and embedding the Poisson singular integral prior to control the smooth degree of image texture. The image degradation model was built based on the high-order statistics technique, and coupled with the priori model to produce the Maximizing A Posteriori. The reconstruction image was obtained by performing the Bayesian inference under the condition of using majorization-minimization principle to solve the MAP. Additionally, optimization of key parameters for image texture restoration with the proposed algorithm was conducted and the user responses of this algorithm were analyzed. In comparison with the current image reconstruction mechanism, the algorithm proposed in this paper had higher restoration visual quality which can better balance the noise and texture. Besides, good user responses of this algorithm was obtained when the image degradation degree was large.