论文部分内容阅读
由于相机异常值严重影响模糊核的正确估计,使传统图像复原算法效果不佳、细节丢失严重、人工痕迹明显,为此本文提出了一种基于消除相机异常值的饱和模糊图像盲复原算法。首先,根据饱和图像灰度特性建立L1正则化模型,添加超拉普拉斯先验提取图像显著边缘。接着,针对S型函数无法完全滤除边缘中的饱和像素,提出一种模糊核镜像辅助函数,通过设定阈值可以有效消除异常值。最后,分析异常值对模糊核估计的影响,建立基于异常值感知的盲反卷积模型,针对迭代求解中二次型问题,采用迭代加权最小二乘法运算得到恢复图像。通过对多幅不同类型的饱和模糊图像进行实验,结果显示复原图像平均灰度梯度高达12.689,图像信息熵达到7.681,处理255*255图像只需6.08s,可以将相机异常值对模糊核估计的影响降到最低,正确估计模糊核函数,保留清晰细节信息的同时显著提高了运算速度,优于当前流行的饱和模糊图像盲复原算法。
Because the camera outliers seriously affect the correct estimation of the fuzzy kernel, the traditional image restoration algorithm is ineffective, the details are lost seriously, and the artifacts are obvious. Therefore, this paper presents a blind restoration algorithm for saturated fuzzy images based on camera outliers. First, the L1 regularization model is established based on the gray characteristic of the saturated image, and the significant edge of the image is extrapolated by using Laplacian prior. Then, in view of the S-type function can not completely filter out the saturated pixels in the edge, a fuzzy kernel mirror image auxiliary function is proposed. By setting the threshold, the outliers can be effectively eliminated. Finally, the influence of outliers on the fuzzy kernel estimation is analyzed, and a blind deconvolution model based on outlier detection is established. For the quadratic problems in iterative solving, the restored images are obtained by iterative weighted least square method. The experimental results show that the average grayscale gradient of the restored image is up to 12.689, the image information entropy is 7.681 and the 255 * 255 image is only 6.08s, which can be used to estimate the fuzzy kernel The influence is minimized, the fuzzy kernel function is correctly estimated, the clear detail information is retained, the computational speed is significantly improved, which is better than the current popular blind algorithm of saturated fuzzy image restoration.