基于运动边界和能量函数最优的图像匹配算法

Images Matching Approach Based on Moving Boundaries and Energy Function Optimization

基于能量函数最优来获取密集视差图(Disparity Map)是图像匹配的新方法。视差图反映了图像中各个景物的深度信息量,因此视差图的边缘并非是对应图像的边缘。而现有方法中能量函数的正则项都依照图像梯度场对初始视差图内部进行平滑并保持边缘的不连续性,这导致了最终获取的密集视差图的平滑区域具有较多的图像边界残留痕迹。光流反映了目标物体的运动状态且包含的运动边界与物体边界相吻合,但光流计算中的正则项导致其在边界上是模糊不准确的,因此采用了多分辨率框架下由粗到精的策略计算大基线时左右图像中的光流场,并联合图像边界和光流来确定准确的运动边界。最后利用所得运动边界构造新的能量函数对视差图进行后处理,实验结果表明无论从视觉效果还是视差图客观评价指标来看,方法都可提取更精准的视差图。

Acquiring dense disparity map based on energy function optimization is a novel stereo matching method. The edge of disparity map which reflects the sceneries＇ depth information is not equivalent to corresponding image edge. But in the current methods, regularization item of the energy function smoothes the interior domain and keeps the boundaries discontinuity of disparity map according to the gradient field of image, and it leads the finial dense disparity map retain much traces of image edge in the smooth area. The motion state of target is mirrored by optical flow among which moving boundaries is identical with the target edge, but the regularization item of optical flow calculation cause the result is inaccurate at the image edge, so the coarse-to-fine strategy in the multi resolution frame was applied to calculate the optical flow for wide baseline stereo pairs, then precise moving boundaries were obtained through the union of image edge and optical flow. Finally, new energy function was defined and post processing was taken to the initial disparity map, the superior visual effect and disparity map evaluation indexes were experimentally verified.