基于样本块修复的目标物移除方法  

Exemplarbased Inpainting of Object Removal

作　　者：季瑞瑞[1] 陈惠兵[1] 许东霞[1] 

机构地区：[1]西安理工大学自动化与信息工程学院,陕西西安710048

出　　处：《现代电子技术》2006年第5期124-125,136,共3页Modern Electronics Technique

摘　　要：结合了常用的整体变分法和图像修补算法的特点,提出了一种从图像中移除不期望的目标物的方法。该方法根据已知的蒙板图像,将图像分为目标区域和源区域。首先计算目标区域边缘象素的优先级,确定目标象素点,然后在图像的源区域中寻找最佳匹配样本块,进行块填充,随着优先级的向内传播,目标物最终被移除,空缺部分被填充为较为理想的象素。实验结果表明,该方法能够有效地移除形状不一的大块目标物,同时对裂痕的修补也有着很高的效率,修复得到的图像自然真实。This paper puts forward a new method to remove the unwanted object from image, combining with the common total variation approach and image inpainting algorithm. According to the known mask,image can be separated into target region and source region. At first,the priorities of the edge pixels in target region are computed to determine the target pixel in each step; then the best matching patch is found in the source region to fill the block around the target pixel, Along with the inward propagation of priority,the object is removed and the vacancy is filled with rather satisfy pixels. Experimental results show that this method can efficiently remove big object in different shape and also function well in image gap inpainting, The result image is natural and vivid.

关 键 词：图像修补 蒙板 优先级 样本块 

分 类 号：TN919.8[电子电信—通信与信息系统]