Directional image analysis with the Hough and Radon transforms

Abstract

The objective of the method to be developed in this paper is to obtain quantitative information on the directional distribution of features in a given image. The Hough transform can be regarded as a method of transforming an image into a domain that reduces the image into a set of parameters from which it is easier to obtain the desired information in the linage. The main drawback of the Hough transform is that it is usually performed on binary images, and is thus dependent on the binarisation method that is used in segmenting the image. The implementation proposed in this paper avoids this drawback by designing a modified version of the Hough transform that uses concepts from the Radon transform to obtain a Hough/Radon transform.

In the proposed algorithm, the image is first transformed into  the Hough/Radon domain, which at the same time a second shadow domain is created to keep count of the number of pixels at a particular parameter point. The Hough/Radon-transformed image L" then filtered to obtain coherent peaks. The shadow parameter space is then integrated to obtain a count of the number of pixels corresponding to directional features in each angle band. Due to the summing of intensities in the Radon transform step, the algorithm detects directional elements on the basis of orientation as well as intensity.

Results with a set of six test images with various types of directional features indicate that the method can perform well over a reasonable range of feature types.