Multimodal stereo correspondence based on phase congruency and edge histogram descriptor

With rapid development in the infrared industry and the availability of relatively cheaper infrared cameras, the simultaneous utilization of visible and infrared data and thus solving the correspondence problem between those images attracts more attention. In this context, we investigate the feasibility of matching features extracted from a multispectral stereo setup using a novel descriptor based on the combination of frequency domain analysis namely “the phase congruency” and spatial information on the shape of the neighboring region around keypoints. A set of stable features are extracted using a bank of Log-Gabor filters at different orientations and frequencies. Features descriptors are computed from the coefficients of the Log-Gabor filters obtained in the previous step and histograms of edge orientations. Matching is based on the cosine similarity function. Experimental results show the performance of this approach to the multimodal correspondence challenge in comparison with other methods cited in the literature.