Unsupervised Method for Change Map Generation

Abstract

Change detection is the process of automatically identifying and analyzing region that have undergone spatial or spectral changes from multi temporal images. Detecting and representing change provides valuable information of the possible transformations a given scene has suffered over time. Change detection is used in several applications (eg. Disaster management, deforestation, urbanization, etc). In this paper a new unsupervised method for change map generation is proposed. Here two multitemporal images are taken as input. In the first step of the approach, absolute-valued difference image and absolute-valued log-ratio image is calculated from co-registered and radiometrically corrected multi-temporal images. These difference images are fused using Discrete Wavelet Transform (DWT). Then, min-mean normalization is applied to the filtered data. The normalized data is clustered into two groups using K-means clustering algorithm as changed pixels and unchanged pixels. To show effectiveness of proposed system, fused image data is given to Principal Component Analysis (PCA) and clustering is done using K-means algorithm. This result is compared with earlier process to show effectiveness of proposed system.