Weighted Multi-Scale Image Matching Based on Harris-Sift Descriptor

Abstract

According to the rotational invariance of Harris corner detectorand the robustness of Sift descriptor. An improved Harris-Sift corner descriptor was proposed. At first, the algorithm given multi-scale strategy to Harris corner, improved corner counting method and removed redundant points at the same time, then, the corner was directly applied to low-pass Gaussian filter image. Based on the histogram of Sift feature descriptor, generates a new 128-dimensional feature vector descriptor by multi-scale Gauss weighted.Through the above, Harris corner detectorand Sift descriptorwas normalizedin the scale layer and gradient features. The experiment results indicated that, the improved corner descriptorcomprised both advantage of Harris corner detection and Sift feature descriptor. The method reduced the computation time and the false match rate, which could be validly applied to the robotstereo vision matching andthree-dimensional reconstruction.

DOI: http://dx.doi.org/10.11591/telkomnika.v11i10.3429