| dc.description.abstract | Image matching is one of the most important processes in Digital Photogrammetry, since it allows the automation of several stages of the photogrammetric pipeline. In most of the commercial software nowadays available, the algorithms of image correlation use the intensity information (gray levels), despising color information, that could be useful, if used in a suitable way, increasing the robustness of the current correspondence techniques in Digital Photogrammetry. The aim of this work is to present a technique that uses the RGB color model in the correlation process, in which a correlation matrix is generated for each color channel. The trace of the covariance matrix related to the translations of the reference window is used to predict which channel can better contribute to the result of the correlation and, with this, to properly weight the correlation coefficients. The weights to be applied to each one of the correlations matrixes are computed adaptively, considering the characteristics of each image. In order to assess this methodology, experiments with real color aerial images were accomplished and correct correlations were achieved with the proposed technique but failed with the current techniques using only grey level images. The results are presented and discussed, showing that the use of color information increases the robustness of the correlation process. | |
