An approach based on texture measures to classify the fully polarimetric SAR image

After intensity value, texture is another significant characteristic which could be used to describe an image. In this paper, a critical analysis based on texture features has been carried out to develop a decision-tree based land cover classification technique for SAR image. The individual role of textural measures like mean, variance, lacunarity, entropy, homogeneity, contrast, dissimilarity, second moment and correlation has been analyzed for examining their ability for classifying SAR image into diverse land cover classes like water, urban, bare soil and vegetation areas. The proposed approach was applied on HV/W polarized image. The concept of separability index has been utilized for analyzing the significance of texture features in separating each land cover class from other remaining classes. Results show that lacunarity proves better for classifying urban from the rest of the classes while contrast and dissimilarity shows better separability for water class. The proposed approach gives an overall accuracy of 84.80%. In future, adaptive land cover classification technique could be developed for classifying and labelling different land cover classes using texture measures.