Detection of curvilinear structures and reconstruction of their regions in gray-scale images

Abstract

In this paper, we present a new method for detecting curvilinear structures and reconstructing their regions in gray-scale images. The concept of skeleton extraction is introduced to detect more general structures such as tapering structures. A candidate skeleton is extracted from the Euclidean distance map that is constructed based on the edge map of an input image. The extracted skeleton is usually noisy due to small protrusions and gaps existing on edge contours. Unnecessary skeletal points are effectively removed with a method combining previously proposed and our own methods. Then, each skeletal point is classified as one of three types ( RIDGE, RAVINE, or STAIR), and connected points of the same type are grouped to form a skeletal segment. Finally, the reconstruction of curvilinear structure regions is performed based on the skeletal segment classification result. Experimental results show that our detector contains many of the desirable properties required of a curvilinear structure detector. Furthermore, since the range of widths that our detector can detect at one time is wide, it is very useful, for example, when an input image includes curvilinear structures of various widths or tapering structures whose width varies greatly. Our algorithm for reconstructing curvilinear structure regions enables us to decompose an image into several types of regions. The reconstruction result, together with the skeleton extraction result, is expected to be useful to make a simplified scene description of an image.