Improved Edge Detection Based On Least Squares Support Vector Machine

Edge detection is the base of the images segmentation, image measuring, pattern recognition and so on. And it is the one of the important subjects in the field of computer vision. It is widely used in the recognition, intensification, segmentation and compress of the image, is often applied to higher-level domain. So, how to detect the edge of object quickly and exactly is a focus of image processing.In recent decades, many scholars work around the edge detection algorithm. In these algorithms, surface fitting method is the relatively good one in recent years. Because of the high accuracy, this algorithm is a more popular method now. Support Vector Machine (SVM) method is the topic surface fitting algorithm.Based on the Least Squares Support Vector Machine (LS-SVM) and multi-scale adaptive Gauss filtering, the new edge-detection method is proposed in this paper. This method not only keeps the excellent performance of method based on LS-SVM in good localization and good detection, but also improves the performance in the detail edge-detection and good detection.Some classic methods and efficient edge-detection methods are introduced in this paper. The basic concepts of these techniques are summarized and the methods advantages and disadvantages are analyzed. Multi-scale adaptive filtering is particularly discussed in this paper. Multi-scale adaptive Gauss filtering is proposed. This method estimates the scale of edge based on local gray scale statistic. The basic theory and the concrete algorithm of the multi-scale adaptive Gauss filtering are summarized.In this paper, the image is filtered with the adaptive multi-scale Gauss filter based on intensity image's mean square value histogram, that is, with the statistical information of the image itself. And the intensity surface of the image filtered by the adaptive multi-scale Gauss filter for the neighborhood of every pixel is well-fitted by LS-SVM, the gradient and the zero-crossing operators are deduced from the LS-SVM with the Radial Basis Function(RBF) kernel. And then the decision is made as to whether a pixel is an edge or not based on the combination results of the gradient and the zero-crossings. Computer experiments are carried out. With the mathematic analysis and test result and Compared with the LS-SVM without using single standard deviation Gauss filter, the proposed algorithm is efficient.