EXTRACTION OF STRIP STRUCTURES USING GRADIENT VECTOR FLOW

Abstract

Extraction of elongated areas, or strip structures in digital images is normally classified as a part of line or curvilinear structure detection. With significant width, the strip structure shows different geometric characteristics from thin linear structures, and makes the conventional line extraction methods impractical in many applications. Another neglected problem in strip extraction lies in the existence of an arbitrary environment, i.e., strips may be bounded by both brighter and darker surroundings in contrast to the assumption in many conventional methods that strips are just dark or bright objects. In the thesis, a novel strip extraction method is proposed by using gradient vector flow (GVF) related computation. It attempts to give a solution to extract strip structures (normally of width larger than 10 pixels) in an arbitrary environment. The method provides a choice of interactive extraction to allow the user to define the end points and an automatic one if features of the strip are known. In the process of the strip extraction, the image is first smoothed to remove noise and fine texture while preserving region boundaries. This is done by using a method based on modified curvature motion. Second, a GVF field is formed from the edge image. A centreline feature image is then obtained from the GVF image. Centrelines of strips are extracted from such a feature image either interactively or automatically. With a minimum distance map to the edge points, an initial contour of a strip gives the approximate position of the strip boundary. Finally, a GVF snake model is used to extract the strip boundary. The algorithm was tested on aerial and micrograph images and good results were obtained.