Urban edge detection by texture analysis

Remote sensing has been a powerful tool to produce structural information to help urban development research and a wide variety of other applications. Nowadays one of the biggest problems of big cities is the lack of information about urban development, and especially the way to obtain fresh urban maps. This work presents a textural approach for edge detection of a urban region. The scheme mixes partial information provided by statistical texture features and a correlation algorithm. The study of gray level co-occurrence matrix (GLCM) and sum and difference histograms (SDH) is used to evaluate textural parameters. High-resolution images showing a urban scene of Mexico City are our main study object. Reached results show that correlation, energy, entropy, max probability and variance features among those proposed by Haralick et al. (1973), produce the most relevant results for used scenes. A faster approach is used to obtain the GLCM, which decreases processing time (Clausi and Zhao, 2001; Morales, 2003). Tested images were reduced from 256 gray-levels to 2 gray-levels using a correlation algorithm. We obtain a good edge extraction; blocks and roads are merged by this method. The algorithms for this work were implemented using MATLAB.