Shape Features Detection Based On Hough Transform In Images

The identification of objects' profile is the most important means to recognize objects by people. It is considerable to detect objects' profile in image processing and pattern recognition. Hough transform as an effective method of shape feature extraction is widely used. But Hough transform is mainly applied on binary image which is also edge image. Therefore before the Hough transform, grayscale images require pre-treatment including image filtering and edge detection. Image pre-processing is an essential preparatory task in the target detection course using Hough transform. Its results will directly affect the quality of the detection. In this paper, we firstly introduce two kinds of noises, which appear in the image commonly. The image filters in spatial domain, Gaussian filter and median filter, which have some limitations, also have been researched in this paper. On the basis of bilateral filtering, we design a bilateral filter based on multi-median sampling of the neighborhood centered at the reference pixel. This method takes care of the distance among the space neighborhood and the similarity of pixels' intensity. The reference pixel value in this filter is obtained by the pseudo-median filtering. This filter can effectively preserve the edge while smooth Gaussian noise and salt-pepper noise and compared to the traditional filters designed for particular type of noise, this method can get good smoothing result for the images with mixed noises (Gaussian noise and salt -pepper noise). And then, a number of classical edge detection operators have been analyzed in this paper. An iterative bilateral filtering method has been introduced to replace the Gaussian filtering process or the adaptive filtering process in the Canny operator, which can avoid blurring the edge to a certain extent in the smoothing process, and get better edge detection results.Hough transform as an effective graphics target detection method can detect straight lines, circles, ellipses, parabolas and many other analytical graphics. The generalized Hough transform has to do some extend to the traditional method, which can detect any graphics using pre-set look-up table and is no longer restricted by graphic analytical expressions. In this paper, we introduce these methods and give the simulation results of these methods. The discretization of image space and parameter space, as well as the calculation of the process make the traditional Hough transform has some limitations, such as poor detection results because of high-intensity noise, a large amount of calculation, large demand of storage resources and so on. This paper analyzes the traditional Hough Transform voting process and points out that the accumulation with 1 in the method is unreasonable. And the traditional methods did not distinguish between the noise points and the points on a straight line in images. So a method of Hough transform based on line connecting tolerance has been proposed in this paper. This method takes account of the overall information of the image and the local information of pixels' neighborhood. So the peaks in parameter space are no longer affected by noises. This method also can avoid the false peaks because of the voting by the feature points with linear relationship. We proposed a Hough transform based on template matching via the modification of the definition of the traditional method. In this method, each parameter unit identifies a template in image space. The feature points according with the conditions can be searched by the template actively. The method takes the number of feature points as the value of parameter unit and takes the record of the coordinates of line segment endpoints. So line segments can be detected and storage resources can be saved.