面向双目立体视觉的迭代式局部颜色校正

在双目立体视觉中,由于相机参数设置差异、环境光照变化、拍摄物体表面非理想漫反射等因素,拍摄获得的立体图像对可能存在颜色差异,进而降低视差计算的准确性.针对这一问题,提出一种面向双目立体视觉的迭代式局部颜色校正方法.首先使用Meanshift算法以不同粒度对2幅图像进行分割,并基于SIFT特征匹配、区域分布及颜色差异初步建立2幅图像中物体间的对应关系;然后使用加权局部颜色校正方法,逐区域进行颜色校正,由于双目图像中物体遮挡范围不同,初步的区域对应存在误差,因此利用校正的双目图像计算视差图,基于视差图像对之间稠密的像素对应,优化立体图像对的区域分割,建立更准确的对应关系,并再次进行颜色校正;迭代地进行立体匹配、优化图像区域对应和颜色校正,直至获得最佳的立体匹配结果.与已有颜色校正方法对基准测试图像集的处理结果表明,文中方法可以有效地提升立体图像对的颜色相似度,提高立体匹配视差结果的准确性.

An Iterative Local Color Correction Method for Binocular Stereo Vision

Due to difference of camera parameters,variance of environmental illumination,non-diffuse reflection of object surface,there is always the color discrepancy between stereoscopic image pair in stereo matching,which will decrease the accuracy of disparity computation.To address this problem,an iterative local color correction method is proposed in this paper.First,the Mean shift algorithm is adopted to segment the stereoscopic images with different granularities respectively.Meanwhile,the SIFT features are extracted and used for local region correspondence initially based on object distributions in the images.Then the target image is corrected by using a weighted local color correction function.Because of different view angles of objects in the images,occlusions will occur and cause the initial region correspondence inaccurate.Thus,a stereo matching algorithm is adopted to generate the disparity maps.The region correspondence is then refined based on the dense feature correspondence between the disparity maps,and the weighted local color correction is performed again.The above refinement will be iteratively performed till the disparity result is convergent.Comparing with several color transfer methods on the benchmark image set,the proposed method can improve the color similarity between stereoscopic image pair and improve the accuracy of stereo matching effectively.