Describe And Match Curve Using Centroid Distance Sequence Based On Frequency Domain

Curve is an important feature of objects in the image, which contains the essential structure of the image, and it provides the most direct information for us to distinguish between the various targets. Therefore, the description of the curve plays a very big role in target recognition. How to describe the curve is an important part of image recognition, and the describing method will directly affect the curve coding and identification results. In the field of machine vision, an ideal curve description method should have the affine invariant including the feature of translation, rotation and stretching of the invariance, and should be insensitive to small perturbation of boundary. If the description method of the curve haves the affine invariant, two curves can be matched further to measure the similarity between the curves. There have been a variety of methods to describe and match the curve, which can be broadly divided into two kinds, known as local characteristics based and global characteristics based. From another point of view, the methods can also be divided into the time-domain based and the frequency-domain based.Contrary to the traditional describing and matching methods, this paper combines time-domain and frequency-domain methods to describe and match the curve. In the time domain, this paper makes some operation on the curve including coarse matching and micro matching of the curve with the global and local characteristics of it. Meantime the paper proposes Fourier transform based on the normalized distance from the centroid of the sequence in the frequency-domain, which can describe the curves in details and make the final match. The selecting of starting point of the conventional curve is more complex, especially when the curve is a closed one. In this paper, the starting point is selected based on the largest distance to the centroid, which is combined with the description method of frequency-domain. When matching two curves, the paper makes use of coarse matching and micro matching to distinguish dissimilarity between curves with distinct difference rather than describe the curves with details first and then judge the similarity between them This measure can improve the efficiency of curve matching to a large extent. Coarse matching and micro matching play a good role in filtering. The frequency domain description method based on the Fourier transform can be a good measure to determine whether the two curves is similar, even if the two curves is not filtered through coarse matching and micro matching.In this paper, experiments are conducted on mpeg7with many pictures. The results prove that the coarse matching and micro matching proposed can play a very good filtering function, and Fourier transform based on the normalized distance from the centroid of the sequence can describe curves well and accurately determine whether the two curves is similar.