Research on edge detection algorithm for plate image

In view of the problems that traditional edge detection algorithms are sensitive to noise and the environment of plate scene is complex, plate image is smoothed with Gaussian filter, and by comparison of edge images from non-subsampled contour let edge detection algorithm and multi-scale wavelet edge detection algorithm, a new algorithm, pulse coupled neural network edge detection algorithm based on multi-scale wavelet transform is proposed. Firstly, multi-scale wavelet is used to detect edge of smoothed plate image, and then pulse coupled neural network is employed to debar the fake edge, followed by binary calculation with K-means clustering algorithm. In the end, after thinning with the modified OPTA algorithm, edges with single pixel-width are obtained. By comparing the experimental results with those of non-subsample contour let transform, it is found that the proposed algorithm not only ameliorates the problem of edge un-continuity caused by non-subsample contour let transform, but the speed is improved.