基于边缘像素约束规则与双特征模型的图像修复算法  

Image inpainting algorithm based on edge pixel constraint and dual feature model

作　　者：高慧[1] 曾庆尚[1] 韩明峰[1] 武栓虎 

机构地区：[1]烟台大学文经学院 [2]烟台大学计算机与控制工程学院

出　　处：《电子测量与仪器学报》2018年第3期32-38,共7页Journal of Electronic Measurement and Instrumentation

基　　金：国家自然科学基金(61170224);山东省自然科学基金(ZR2012FL07)资助项目

摘　　要：为了解决当前基于样本块的图像修复算法主要是常采用固定尺寸的样本块进行重构,在面对丰富的边缘细节信息时,导致重构图像中出现振铃效应以及不连贯效应等不足,提出了一种基于边缘像素约束与双特征模型的图像修复算法。首先,将图像信息熵引入Criminisi算法中,改进了优先级计算函数,确定图像修复的顺序。然后,利用Sobel算子对图像进行边缘检测,利用边缘像素所对应的灰度信息来建立边缘像素约束规则,通过边缘像素点的灰度值之差动态选择合适尺寸的样本块。最后,通过联合SSD和SSIM模型,设计双特征模型,对待修复块以及匹配块进行相似性度量,确定最佳匹配块对损坏区域块进行填充,完成图像修复。实验结果显示,与当前图像修复算法相比,所提算法具有更高复原质量与效率。所提算法具备较好的图像修复能力,在图像处理领域具有良好的参考价值。At present,many image inpainting algorithms usually use image samples with size invariance to repair images in the process of image inpainting When there is a rich edge detail information in the image to be repaired,the ringing effect and the incoherence effect are not enough,which makes the algorithm less robust and less effective in restoration. In this paper,an image inpainting algorithm based on edge pixel constraint and dual feature model is proposed. Firstly,the image information entropy,which measures the amount of image information,is introduced into the calculation of the priority of the pixels to be repaired,and the order of image restoration is determined by the calculation of the priority of the pixel points. Sobel operator is used to detect the edges of the image. The edge pixel constraint rules are established by using the gray information corresponding to the edge pixels. The appropriate size samples are selected.Finally,a double feature model is designed by weighing and constructed the SSD model and SSIM model,treating repaire block and matching block with similarity measure,and filling the best matching block with the damaged area block to complete the image restoration. Experimental results show that the proposed algorithm has stronger robust performance and better repair performance compared with the current image inpainting algorithms. The proposed algorithm has better image restoration capability and shows excellent reference value in image processing field.

关 键 词：图像修复 优先级计算 SOBEL算子 边缘像素约束规则 双特征模型 最佳匹配块 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]