Detection and feature extraction method of curvilinear convex regions with weak contrast using a gradient vector distribution model

We recognize a region of almost rectilinear high intensity mound as a kind of lines. This shape is very important to understand image. But there are only a few general methods to detect such `line'. In addition, most of these methods have two major problems. One is that the performance of edge detection methods depends severely on the noise conditions. The other is that it also depends on the contrast between the line and its background. Because of these two problems, it is difficult to detect lines with various contrasts in real images reliably. In this work, we propose a new filter to detect and enhance such lines. It is robust against noise disturbances and also its output does not depend on the contrast. We first define the line-convergence vector field model based on the distribution of gradient vector orientation. Next we define a criterion index called the line-convergence degree to evaluate the likelihood of the existence of a line. The output of the proposed filter is defined as the average of line-convergence degrees in a region which is adapted to the gradient vector distribution. The filter output is a function of only gradient vector orientation and it is free from absolute intensity and relative contrast variations. Experimental results using artificial images and real images show the effectiveness of the proposed filter.