基于修正点扩散函数的超分辨率复原算法

Super Resolution Restoration Based on Modified Point Spread Function

针对传统的超分辨率复原算法边缘保持能力不足,存在振铃效应等问题,提出了基于修正点扩散函数的凸集投影超分辨率复原算法。首先,检测参考图像的边缘;然后,对传统的点扩散函数加一个权值因子进行修正,将点扩散函数分为0°、22.5°、45°、67.5°、90°、112.5°、135°、157.5°8个方向,达到在边缘部分降低点扩散函数作用范围的效果;最后,利用改进的点扩散函数迭代修正参考帧,直到估计灰度值与实际灰度值的误差小到一定范围或达到设定的迭代次数,退出迭代,得到超分辨率复原图像。复原图像的质量采用峰值信噪比、均方误差和结构相似度进行评价。实验结果表明,两类测试图像的峰值信噪比提高范围为3.46~6.91 d B、均方误差降低范围为43.47~87.82、结构相似度提高范围为0.050 8~0.381 7。提高了超分辨率复原的边缘保持能力和复原图像的质量。

Because the traditional super resolution restoration algorithm edge preserving ability is insufficient and the ringing effect,we propose the projection on convex sets super resolution restoration algorithm based on modified point spread function. First detect the edge of the reference image; Then,modify the traditional point spread function using a weight factor,and the point spread function is divided into the eight directions of 0°,22. 5°,45°,67. 5°,90°,112. 5°,135°,157. 5°,to achieve the effect of reducing the range of point spread function in the edge section; Finally,modify the reference frame iteratively using the improved point spread function,until the error between the estimated gray value and the actual gray value is small to a certain range or the set iteration number has been reached; Exit iteration and get super-resolution restoration image. The quality of restored image is evaluated by the peak signal to noise ratio,mean square error and structural similarity.Experiment results indicate that the peak signal to noise ratio of the two types of test images is improved by3. 46 - 6. 91 d B,the mean square error is reduced by 43. 47 - 87. 82,and the structural similarity is improved by 0. 050 8 - 0. 381 7. It is concluded that the proposed algorithm improves the edge preserving ability of super resolution reconstruction,and improves the quality of the restored image.