Descriptors based Unsupervised Change Detection in Satellite Images

Abstract

In this thesis, a novel unsupervised technique is proposed to get the change analysis of bitemporal satellite images. The proposed technique is based on combination of Multi-Block Local Binary Pattern (MB-LBP) descriptor and Binary Robust Independent Elementary Features (BRIEF) descriptor. In this approach, the MB-LBP encodes overlapping blocks of both images based on BRIEF encoded decimal values of square sub-regions of considered overlapping block. BRIEF generates decimal values by comparing each pixel in each sub-region with a calculated quantity. This quantity is generated by taking an average of both center pixel and a calculated threshold. To get binary feature vectors for each corresponding pixels of both images, MB-LBP is applied on the obtained decimal values. Hamming distance is used as a similarity metric to compare the binary feature vectors of each pixel of both images. This provides binary change map of changed and unchanged region. Optical satellite images acquired by Landsat satellite are used to perform the experiments. Experimental results show that the proposed method provides better results compared to earlier reported techniques like expectation maximization and kernel k-means methods.