Color-Texture Analysis To Improve Sentinel Urban Image Classification

This paper investigates the using of multiband compact texture unit descriptors for intra-band and inter-band texture analysis. The extracted features are used in a patch-based classification of optical Sentinel-2 and Radar and Sentinel-1 images over an urban area to discriminate between different housing types. To evaluate the pertinence of the multiband texture features, various state-of-the-art color-texture feature extraction methods, such as Integrative Grey Level Co-occurrence Matrices (IGLCM), Opponent Gabor (OGabor) filters, Opponent Local Binary Patterns (OLBP), are considered and compared by means of K-nearest neighbors’ classifier. Results from this study show that proposed approach produces the best accuracies, with the lowest computational time