一种改进的Criminisi眼底图像修复算法

An Improved Criminisi Fundus Image Restoration Algorithm

为了抑制眼底图像中的光斑干扰,改善其图像质量,在Criminisi算法的基础上,通过优化亮度局部方差、引入亮度局部绝对值以及梯度相似度与亮度相似度,提出了一种新的眼底图像修复算法。为了增强算法鲁棒性,将经典Criminisi算法中的优先权公式由相乘改为相加;将亮度局部绝对差引入优先权计算公式中,提高优先权的合理性,利用亮度局部绝对值自适应地选择样本块的大小;引入梯度相似度与亮度相似度,修改了算法中图像块的匹配算法。通过与经典Criminisi算法以及其他同类改进算法对比,本文所提算法修复效果在修复区域与完好区域之间的过渡更加自然,修复痕迹更少,且更符合人眼的直观视觉感受与图像的构图特征;同时,峰值信噪比相比传统的Criminisi算法平均提高了2.0 dB,结构相似性与其他算法相比也均有所提升。

In order to suppress the spot interference in the fundus image and improve its image quality,a new fundus image restoration algorithm was proposed by optimizing the local variance of brightness,introducing the local absolute value of brightness,and introducing gradient similarity and brightness similarity based on the Criminisi algorithm.In order to enhance the robustness of the algorithm,the priority formula in the classical Criminisi algorithm was changed from multiplication to addition;the local absolute difference of brightness was introduced into the priority calculation formula to improve the rationality of priority,and the size of sample block was adaptively selected by using the local absolute value of brightness;the gradient similarity and brightness similarity were introduced to modify the matching algorithm of image blocks.Compared with the classic Criminisi algorithm and other similar improved algorithms,the restoration effect of the proposed algorithm in this paper has a more natural transition between the repaired area and the intact area,with less repair traces,and is more in line with the intuitive visual perception of the human eyes and the composition characteristics of the image.At the same time,compared with the traditional Criminisi algorithm,the peak signal to noise ratio is increased by 2.0 dB on average,and structural similarity is improved compared with other algorithms.