Symmetry Detection In Vision Description

The detection of symmetry feature in visual feature description is emerging as an active research area involving several subjects such as image processing, pattern recognition, computer vision, physiology, psychology, and so on. The detection of symmetry is always the important content of understanding of the image, and computer vision. Symmetry feature can influence the mechanism of human's notice so that can instruct the detecting, selection and recognition of the object in human's evening vision. Symmetry feature is also treated as important parameter in physical, chemical research and medical diagnoses. With the fast development of the retrieval of the image based on content, the technology of the detection of the symmetry becomes perfect step by step, so that it can be applied in sweeping areas.This paper focuses on the principal problems in symmetry detection such as the selection of the symmetric points in image and the location of reflectional symmetry axes and deeply research in them. Firstly, by imitating human brains' visual characteristic and applying the information of the phase, we find a measure of symmetry base on wavelet transform computed by phase congruency function, based on which we can choose the areas of symmetry in images. Secondly, in order to find the location of symmetry axes in the images, we compute the orientation histogram of the local energy gradient. Lastly, we extend the method to deal with an order of images with symmetric object. The results of the experiment can find different diatoms images with different orientation and number symmetry axes, and retrieve them by symmetry detection. We estimate detection of symmetry has a broad practicability.The software in the paper is programmed with Matlab 5.2,debugged successfully on IBM PC 586(400MHZ). The used images in the paper are 8 bit gray images. The method in this paper could be well applied in the application of retrieval of the images based on content.