Dense correspondence and optical flow estimation using gabor, schmid and steerable descriptors

Abstract

In this paper, the use of three dense descriptors, namely Schmid, Gabor and steerable descriptors, is introduced and investigated for optical flow estimation and dense correspondence of different scenes and compared with the well-known dense SIFT/SIFTFlow. Several examples of optical flow estimation and dense correspondence across scenes with high variations in the intensity levels, difference in the presence of features and different misalignment models (rigid, deformable, homography etc.) are studied and the results are quantitatively/ qualitatively compared with dense SIFT/SIFTFlow. The proposed dense descriptors provide comparable or better results than dense SIFT/SIFTFlow which shows the high potential in this area for more thorough investigations.