采用多方向插值融合的快速铁路货运图像修复

Fast Inpainting of Railway Freights Images Based on Multiple Direction Interpolation

由于非目标前景的出现,高分辨率的线阵相机采集到的图像常因表面原始信息被遮挡而产生数据缺失.针对单张图像提出一种基于多方向插值融合的图像修复方法,并成功应用于铁路装载状态检测系统中,实现了监控图像中高压线的去除.在获得图像中待修复区域的位置后,将待修复区域分为平滑区和边缘区.对平滑区域,在3个不同角度方向上选择梯度变化最小方向上的邻域像素,用三次插值算法对待修复像素进行估计;对边缘区域,结合3个方向的三次插值结果进行数据融合,设计融合规则以获得更准确的插值结果.以HSI色彩空间代替RGB色彩空间对彩色图像进行修复.与BSCB、FMM等算法进行对比,表明所提出的算法能获得更好的修复效果,并能有效修复色彩复杂的图像,同时还降低了计算复杂度,提高了运行效率,使处理时间达到毫秒级,可满足实时处理要求.

High resolution line-scan cameras produce images with missing information due to occlusion caused by foreign objects. In this paper, we propose a framework based on multiple direction interpolation for single image digital restoration to remove high tension wire （HTW） from images for railway loaded condition inspection （LCI） applications. The known region to be reconstructed is automatically segmented into smooth and edge sub-regions. The smooth regions are filled-in with bilineax interpolation based on the information in the neighborhood, which is the minimum among the gradient values in three directions. For the edge regions, the fill-in result is optimized using a data fusion method based on the information in three directions with bi-cubic interpolation. Inpainting is performed for each channel independently based on the HSI color space instead of RGB. Experimental results indicate that better restoration results can be obtained as compared with traditional Bertalmio-Sapiro-Caselles-Bellester （BSCB） and fast marching method （FMM） algorithms. The proposed algorithm can also effectively deal with inpainting of complex color images with high computational efficiency.