Urban and non urban area classification by texture characteristics and data fusion

Over the past few years, an increased need has been observed in order to obtain high resolution data. Aerial photographs have demonstrated potential to provide complementary information for a wide variety of applications. In this context, our paper presents a study to classify urban and non urban regions of panchromatic images. The scheme takes into account the partial information provided by statistical textural features. Based on a data fusion algorithm, textural data are merged and a binary classification result is obtained. The study implements Gray Level Co-occurrence Matrices (GLCM) to evaluate textural parameters applied to high resolution images showing an urban scene of Mexico City. Reached results showed that energy, correlation and cluster shade features among the 14 originally proposed by Haralick et al., were the most relevant results for the used scenes. Co-occurrence matrices algorithm requires a lot of computing processing, thus a new approach is proposed which decreases computation time. The tested images with 256 gray-levels were reduced to 64 gray-levels. By means of an optimal correlation algorithm, results from texture were reduced to 2 gray-levels to take place in the fusion process. The fusion process use error measures and map knowledge to determinate the final result. Results obtained by this method show that a good classification of streets, blocks and non urban regions can be made. The algorithms for this work have been implemented using MATLAB.