Framelet transform based edge detection for straight line detection from remote sensing images

Edge detection has been widely used as a pre-processing step for image processing applications such as region segmentation, feature extraction and object boundary description. Classical edge detection operators available in literature are easy to implement, but not all the edge detection operators is suitable for remote sensing images in terms of selecting threshold and kernel function. There is no acceptable method to select the parameters in classical edge detection methods. Multiresolution analysis such as wavelet transform has been shown to have advantages over classical edge detection techniques, as it is less sensitive to noise. The discrete wavelet transform (DWT) is shift variant, due to critical subsampling. The DWT is not capable of capturing edges, which are not aligned in horizontal and vertical directions. In this paper, we focus beyond DWT, framelet transform used to detect edges from LISS III and Cartosat images. The proficiency of the proposed method is evaluated by comparing the results of DWT, dual tree complex wavelet transform (DTCWT), curvelet transform (CUT), contourlet transform (CT) and non subsampled contourlet transform (NSCT) based edge detection methods. Rosenfeld evaluation metric is used to measure the quality of the edge detection methods, which shows the framelet based edge detection produce sound results than other methods. Principal component analysis (PCA) and singular value decomposition (SVD) methods are used to remove the correlation among the multispectral bands and selected maximum information bands for edge detection, instead of using one particular band because each band in multispectral image is suitable for specific applications. The detected edges are further subjected to line detection algorithms such as standard Hough transform, small eigenvalue analysis and principal component analysis. The outcomes are compared in terms of complexity measurements. Framelet transform along with principal component analysis based line detection algorithm perform better than other two methods.