Line Segment Detection by High Performance Digital Template Hough Transform DTHT4

We are not free from the side ef'fect caused by the digitization of the image also in Hough transform. Since Hough transform detects lines in the expression of the analog line equation, it is finally needed to extract digital line segments from the analog line equations. In this paper, we propose a new method DTHT(Digita1 Template Hough Transform) to detect digital line segments directly from the digital edge image. The basic idea of DTHT method is that the digital template of line segment can replace the template line equation in the conventional Hough transform. DTHT can be regarded basically as the new Hough transform with 4-dimensional parameter space indicating a couple of two end points of the segment. DTHT algorithm can directly detect line segments without post-processing of Hough Transform. DTHT can perform about 30 times faster than the original t-lough transform, and DTHT is in~portant both in theoretical and practical aspects. Hough Transform is one of the most useful methodologies for detecting lines from noisy real images. On the other hand, Hough transform has been inheriting several kinds of unsolved problems caused by the digital implementation of Hough transform on the digital image : quantization problem in the Hough parameter space, threshold selection problem for peak extraction and line segmentation problem from line equation. This paper proposes a new Hough transform method that can put these fundamental problems into consideration simultaneously. This new method is called Digital Template Hough Transform, DTHT, and DTHT can be regarded basically as a Hough transform for line segment detection with 4-dimensional parameter space. Since DTHT algorithm extracts edge line segments by means of the direct matching of digital edge image with the digital template of line segment, DTHT algorithm needs no requirement for preparing for the quantization problem of the parameter space, no requirement for tuning the threshold for peak extraction and no requirement for the segmentation of the line. In spite of the superior properties of DTHT, DTHT would be suff'eretl mainly from the increasing of compirtation cost and the increasing of the complexity for the detectabilities of line segments. In this paper, we present DTHT algorithm in Chap.3 basing on the considerations on the basic properties of Hough transform in Chap.2. Additionally we propose a few countermeasures to decrease the computation cost and the complexity lor the detectabilities. It was experimentally tlcn~onslrated