Improved Fourier-mellin The Descriptor Image Matching

In recent years, Fourier-Mellin descriptor has been widely applied in digital image matching. But because of the noise of spectrum aliasing and errors, Fourier-Mellin descriptor can not be sure that the form of the changed relative value distribution is the standard two-dimensional pulse function.Based on the basis of introducing distance formula with kernel function, first using wavelet transform in frequency domain extract stable feature points to form feature vector.Because of wavelet transform has local analysis ability and analytical skills, it can effectively extract to constant translation, rotation, dimension transformation and small deformation of image of effective features, wavelet transform is widely used in the extraction of characteristic points. Using wavelet transform can get horizontal and vertical and diagonal three edge profile in the frequency domain, The stayed matching image after wavelet transform is decomposed into son four band with the HL,LH LL bands, and HH band through comprehensive filtering, interpolation, stacked single sub-band reduction into stayed matching image of the original image. In reduction process,if HL,LH and HH these three bands of edge feature points can overlap together, then these feature points are stable feature points.Then average displacement algorithm is used to amend feature vector, if kernel function is meeted certain conditions, the stayed matching image estimate is consistent and unbiased. For the Epanechniov kernel functions and properties of monotonic increasing kernel function,average displacement method has better convergence, this convergence in this paper is demonstrated. The feature space can be clustered by employing the average displacement algorithm. Finding the peak value of the distribution function can be achieved by searching the cluttering center,thus making the formation of standard two-dimensional pulse function.Average displacement algorithm is an adaptive rising fast algorithm, it for any one point to compute the average displacement vector, calculate later, turn search window namely nuclear in the direction of the vector pointing constantly on translation, in every translation get a new position, in the new position repeat this process, through such repeated mobile, has been moved to reach the nuclear center namely peak.Through this paper improvement,to correctly compute matching parameters, so as to achieve the matching result of two images. The experimental results show that the new algorithm is efficiency and reliability.