Convergent Floating Thresholding for Image Extraction

Abstract

In this paper, the authors propose a method for thresholding a gray-scaled image. In the method, a threshold for a coordinate is set by an iteration that is composed of the pixels detection from neighboring area and a calculation of a candidate threshold using the detected pixels. The candidate threshold settles down to a value, because the detected pixels finally become invariable. Therefore, the method can set the threshold which follows up a light changing and a contrast changing. The proposed method and the conventional methods were tested, using a preceding car images of the inside and outside of a tunnel. We compared the method with other methods by an estimation based on a correlation on the thresholding. The result showed that the proposed method was stable relatively and set the good threshold continuously.