基于增强型K-NN块搜索的图像修复算法

Image inpainting algorithm based on enhanced K-NN search

针对当前图像修复方法在对遮蔽物损坏图像复原时,存在明显的模糊效应与不连续效应等不足,提出局部最小二乘逼近优化耦合增强K-NN块搜索的图像修复算法。通过对图像修复机理进行分析,联合等权方法与K-NN(K近邻)块,将未知像素的估值转化为对线性组合函数的求解;定义基于边缘的优先项,计算输入块的边缘特性,提出基于局部学习映射函数的增强型K-NN块搜索方法,降低未知像素值K-NN的误配;采用基于局部最小二乘逼近优化方法,将相似块中的像素传播至损坏区域,完成图像修复。测试结果表明,与当前图像修复算法相比,在遮蔽物损坏图像复原中,该技术拥有更好的修复质量,有效降低了模糊效应,克服了修复时存在的间断效应。

To solve these drawbacks such that when using the current image inpainting algorithm processes the image damaged by concealment,the obvious blurring effects and discontinuous effects emerge in the repairing areas,an image inpainting algorithm based on local least square approximation optimized and enhanced K-NN search was proposed.By analyzing image repair mechanism,combined uniform weights with K-NN patches,the estimation of unknown pixels was obtained by computing a linear combination.Edge-based preferred term was defined,and the term was taken into account.The edginess of the input patch was calculated,an enhanced K-NNsearch method was proposed based on local learning of mapping functions,to cope with that the found K-NN might not correspond to the unknown pixels.The algorithm based on local least square approximation optimization was used to spread pixels in domain blocks to damaged region,to complete the image inpainting.The simulation results show that comparing with current image inpainting algorithm,this algorithm has better inpainting quality and effectively reduces the blurring effects,overcomes the discontinuous effects in inpainting of images damaged by concealment.