Histogram‐based cost aggregation strategy with joint bilateral filtering for stereo matching

Abstract

The edge‐aware bilateral filter has been demonstrated to be effective for preserving depth edges, and disparity maps obtained from Fast Bilateral Stereo (FBS) have enhanced the efficiency of algorithm and the robustness to noise. However, they also lead to a non‐perfect localisation of discontinuities. To overcome this issue, a new bilateral filtering based cost aggregation utilising colour statistical classification and similarity measurement within annular blocks is proposed in this study. We have adopted the similarity of histograms evaluated by Earth Mover Distance (EMD) to obtain the raw matching cost in the raised annular block, since histograms are very effective and efficient in capturing the distribution characteristics of visual features. For the weights aggregation, the spatial weight is assumed to be a constant. The colour weight is calculated by using a cluster‐mean‐value strategy, which is implemented by the local colour histogram. It improves the accuracy in the discontinuous areas. Computation redundancy is reduced by disparity candidate selection using the local minimal relevancy in the corresponding annular blocks. We use the efficiency and accuracy to demonstrate the performance of our proposed method. Experimental results have shown that the proposed method reduces the mismatch at depth discontinuous and the computation complexity significantly.

Keywords

• histogram-based cost aggregation strategy
• stereo matching
• edge-aware bilateral filter
• bilateral filtering based cost aggregation strategy
• colour histogram statistical classification
• similarity measurement