【摘要】 针对现有非局部立体匹配算法在不同纹理图像下因选取固定参数而导致整体误匹配率偏高的问题,提出一种自适应参数的非局部立体匹配算法。对颜色差和水平及垂直方向梯度代价进行归一化处理并加权融合,通过对融合后代价作对数变换,提高了单像素匹配代价在低纹理区域表现;在色度,饱和度,纯度(HSV)颜色空间上,计算参考图的平均色度差,自适应调节代价聚合参数;经过视差选择获得视差图。实验结果表明:所提算法在Middlebury测试平台对31组立体图像对处理后的平均误差为7. 86%。在低纹理区域误匹配率进一步降低,具有更好的泛化性。更多还原

【Abstract】 Aiming at problem that current non-local stereo matching algorithm has high false matching rate due to the selection of fixed parameters under different texture images,a non-local stereo matching algorithm with adaptive parameter is proposed. Color difference and cost of horizontal and vertical gradient are normalized and fused through weighing. Then a logarithmic transformation is performed based on the fused cost,which improves the performance of single pixel matching cost in low texture area. The average chroma difference of the reference image is calculated in hue,saturation,value( HSV) color space,and then adjust cost aggregation parameter adaptively.The final disparity map is obtained after disparity selection. The experimental result show that the average error of the algorithm is 7. 86 % on the Middlebury test platform after processing on 31 groups of stero images pair. The mismatching rate in low texture region is further reduced,and the generalization ability of proposed algorithm is much better.更多还原