A fusion-based segmentation algorithm for high-resolution panchromatic aerial photography

We propose a mixed segmentation algorithm based on both gray level information and a texture parameter. The definitive k-class image is obtained by means of a simple fusion scheme. Our algorithm considers the following steps: (a) obtainment of an n-class image by means of an algorithm based exclusively on spectral properties, (b) obtainment of a texture image, which may be generated by Markov random fields (MRF) modeling or by means of the co-occurrence matrix. (c) From the n-class image of the first step we obtain a mask for each class; based on these masks we analyze the variation degree of the texture parameter: if the variation degree is greater than a defined threshold the class is fissioned into 2 classes, otherwise it remains the same. (d) The definitive class image is obtained by fusing the different sub-images created in the previous step. Results show that considering complementary information in the segmentation process results in a better discrimination among classes, while the main edges of the scene are clearly defined.