Robust sub-pixel disparity estimation and its refinement around depth discontinuity and featureless areas

This paper presents an enhanced robust phase correlation (ERPC) method for sub-pixel disparity estimation together with a package of refinement techniques for DEM generation. The multi-scale ERPC algorithm can cope with very large disparity measurement on the one hand and improve sub-pixel accuracy by entirely avoiding the ill-posed problem of 2D phase unwrapping on the other. Recognizing the inherent weakness of phase correlation as an area matching based method, we then present a refinement scheme with compound phase correlation (CPC) and Median Shift Propagation (MSP) filter techniques to deal with the unreliable disparity estimates around the depth discontinuities, featureless areas and other low correlation areas. Through this refinement scheme, we are able to greatly improve the accuracy of phase correlation based disparity estimation for DEM generation from stereo image pairs with versatile baseline settings (from narrow to wide baseline).