High Accuracy Disparity Estimation via Weighted Phase Correlation Disparity Tracker

Abstract

When using Phase Correlation (PC) to estimate disparity map of remote sensing images with narrow baseline, disparity precision is usually influenced by matching window size, low texture areas and motion noises. To solve these problems, we propose a weighted PC based disparity tracker: First, a hierarchical meth-od is applied to put images divided into multistage structure; Then, at first level, weighted PC employs initial window to esti-mate a rough disparity matrix. Meanwhile, Kronecker delta peak of initial window is applied to get a reliable threshold. Next, disparity tracker employs three sub-steps to refine the disparity matrix: first, window size is reduced according to adaptive win-dow strategy; then, in target image, the reduced window is moved to a new position under the guide of prior disparity ma-trix; third, weighted PC is employed to estimate the disparity matrix of next level. Finally, the disparity tracker is iteratively executed until it converges. Moreover, according to the reliable threshold, a reliability evaluation approach is imbedded into disparity tracker for the reliability of each disparity. Comparing with the state-of-the-art PC methods, the experimental results demonstrate our method obtains better results in visual and quantitative evaluations, especially in minimizing the influence of tremendous depth differences and motion noises caused by high mountains and river areas.