REGION SEGMENTATION AND CLASSIFICATION OF HIGH RESOLUTION IMAGERY

摘要

Segmentation and classification of high resolution imagery is a challenging problem due to the fact that it is no longer meaningful to carry out this task on a pixel by pixel basis. The fine spatial resolution implies that each object is now an aggregation of a number of pixels in close spatial proximity, and accurate classification requires that this aspect be considered. A solution for this problem is to segment the image into a collection of regions based on similarity / dissimilarity and then classify the regions using both multispectral and spatial attributes. We describe an approach here wherein the input 1-metre resolution image is first segmented using the morphological watershed transform to extract the regions. A number of properties of the regions are computed – spectral mean vector, average texture, departure from circularity, length-to-breadth ratio, area, perimeter, compactness and others. Small regions are merged with the most similar neighboring regions based on a user specified threshold on the size. Where the image is dominated by texture, texture features based on gray-level co-occurrence matrix are generated and can be used to extract the regions. After segmentation the image is classified on the basis of the regions instead of the pixels. Where the pixels were of the order of a few millions, the regions were of the order of a few thousands. These regions are classified using artificial neural network. The results are encouraging and the scheme developed in this study is being evaluated with a range of images and a number of other classifiers.