Application of the linear feature detection system—LINDA to image segmentation from remotely sensed data

Abstract

Abstract In this study the linear feature detection and analysis system—LINDA is applied to image segmentation using both structural and spectral information from remotely sensed data. A new procedure for image segmentation is proposed which involves the following steps: (1) Apply LINDA's modules “Line Detection” or “Edge Detection” to extract linear feature networks from remotely sensed data; (2) Apply LINDA's module “Structural Measures” to obtain a network density map; (3) Image segmentation based on network density; (4) Image segmentation based on brightness values of original data; (5) Combine the results from Steps 3 and 4; (6) Post‐processing. An example is given for urban‐rural area segmentation using a Landsat TM image of Kitchener‐Waterloo area in Ontario, Canada. The result is satisfactory. It demonstrated that the performance of image segmentation from remotely sensed data can be improved by utilizing both structural and spectral information, instead of using one type of the information alone. It also demonstrated that the LINDA system is a useful tool for remote sensing image segmentation.