Shape Recognition Based On Fourier Descriptor Of Moments

Shape is the inherence characteristic of an object, and it is the important character used for the object recognition. So the shape recognition of the object has become a very important research direction in Computer Vision and Pattern Recognition, which has been widely applied to areas such as product testing and target tracking.Feature description of shape is the key of image recognition, and the quality of feature extraction directly determines the effect of the target recognition. So feature description and extraction of shape is the main content of this topic.This article focuses on the moment invariants and Fourier descriptors, which are the two kinds of classical methods in feature descriptions of shape. Their basic principles, methods and some typical improved algorithms are discussed and analyzed in-depth, and their respective advantages and disadvantages of two types of algorithms are summarized.To apply to real-time image processing of target recognition system, a new method for shape recognition in which moments and Fourier descriptors were combined was presented. The object was divided into a sequence of fan-shaped sub-regions, originated at the centroid of the object image. The moments calculated in all the sub-regions formed the moment sequence to represent the characteristics of the object. The Fourier Descriptor of the moment sequence was then derived using DFT, which has properties of invariant in the translation, rotation and scale transform.The experiments results demonstrated that this technique has properties of invariant in the translation, rotation and scale transformation of target shape, can identify the shapes of 2-D objects accurately, as well as for 3-D objects, and has very good real-time performance and high recognition rate. It shows this new method has very good outlook engineering application.Compared with traditional moment invariants of shape, this method has a small amount of computing, faster processing speed, and can meet the requirements of real-time image processing. Compared with traditional Fourier descriptors, this method has better stability, stronger ability to distinguish between, and higher recognition rate.