针对当前局部立体匹配视差精度低的问题,提出了一种二次引导滤波模型,并应用于局部立体匹配算法。在新设计的二次引导滤波模型中,将第一次引导滤波的输出图像作为第二次引导滤波的引导图像,克服了传统引导滤波的缺陷,抑制了噪声。在代价聚合阶段引入二次引导滤波,使用跨尺度框架聚合各尺度的匹配代价,进一步提高算法匹配精度。实验结果表明,基于二次引导滤波的局部立体匹配算法在Middlebury测试平台上对标准立体图像对的检测具有更高的精度,且代价聚合步骤的时间复杂度与滤波窗口大小无关,在匹配精度和速度上都取得了良好的效果。二次引导滤波的思想有望在立体匹配等领域取得更广泛的应用。

To solve the problems of low disparity accuracy in local stereo matching, a secondary guided filtering model is proposed and applied to the local stereo matching algorithm. The newly designed secondary guided filtering model overcomes the deficiency of traditional guided filtering and further suppresses the noises because the result of the first guided image filtering is used as the guiding image of the second guided filtering. In the cost aggregation phase, the introduction of the secondary guided filtering further improves the matching accuracy because the cross-scale framework is used to aggregate the matching cost volume at each scale. The experimental results demonstrate that the local stereo matching algorithm based on the secondary guided filtering possesses a high accuracy in the detection of standard stereo image pairs on the Middlebury benchmark. Moreover, the temporal complexity of the cost aggregation phases is independent of the filtering kernel size, and the proposed algorithm achieves good performances in speed and accuracy. The idea of the secondary guided filtering has potential applications in stereo matching and others.