Improved Algorithm Of Edge Detection Based On B-spline Wavelet

The paper reseaches the algorithm of edge detection on wavelet for the purpose of the best compromising between removing noise and remaining fine edges.At first, the background and the actuality of image edge detection are introduced, basic concept and characteristic of edge detection are dissertated, and then the typical algorithms are summarized. They can't work well with some noisy images as well as top-quality images.Next, edge detector based on wavelet transforms is chifelly analized. The development of wavelet analysis and its fundamental theories are introduced as well as multi-scales wavelet transforms is proposed. But the proposed approach will cause displacement and loss of edges. Therefore, we put forward that the proposed approach can be combined with the Canny detector according to the ideal of data fusion based on wavelet transforms.At last,the criterion of image edge detection is introduced.In summary, three aspects work is carried out in the paper of the following:(1)for the gray-scale image of higher resolution (1024×1024 standard image), there are two important problerms in B-spline wavelet edge detector:①The algorithm of multi-scale fusion will let the faint edge disappear;②edges have been extracted from a certain scale are not retained .They continue to be smoothed and displacemented. It reduces the resolution of the image and wastes the time of computation.Addressing these two issues, improvements are made as follows:①The original algorithm"the whole image do wavelet transform in the next scale– the image is judged by threshold value - Multi-Scale Integration"is changed,"Edge is judged by threshold value firstly- edges have been detected are directly to retain, that not detected do next-scale wavelet transform"is used in the paper. The original algorithm can be improved with this method to a certain extent:(a)The algorithm improved has a reduction of multi-scale integration from the original algorithm, so faint edges can be effectively saved.(b)images are judged firstly by threshold value, pixels in line with the conditions no longer do next convolution, deconvolution on the whole image of the original algorithm is replaced, multi-scale integration is reduced, all of them save the computation time.②The initial wavelet transform is replaced by Contourlet transform to set a"adaptive threshold"to judge Edge and noise.③The final outcome is thinned using morphological to get single-pixel edge.(2) for low-resolution images, in view of the inherent characteristics of Gauss:its space coefficient does not have the function of automatic adjustment that once value is defined, it Smoothes images do not vary due to differences of signal to noise ratio, an mult-levels adapive space coefficients edge detector based on Gauss wavelet is put forward to in the paper.LOG is improved by many algorithms, the algorithm that image is blocked, with the adaptive design of the correspondingσby region is put forward to, but there are two major shortcomings:①complexity of the algorithm. Calculating block adaptiveσis of the overload.②Laplacian algorithm as to the superiority of their ownself is poor than wavelet detection, why improved algorithm LOG effect is not satisfactory.Firstly, inertia of moment of gray level co-occurrence matrix is used to design theσwhich is suitable to the current image. Secondly, high-pass filters and low-pass filters are designed according to theσand the nextσis determined out in accordance with the image which is filtered by low-pass filters. The process is repeated till noise is removed basically. At last, images of all levels extracted by differentσare fused to obtain the final image edge with only one pixel wide in accordance by certain rules.(3)According to the problem that exacting edge only in the high-frequency part without low-frequency information would lose details, an algorithm combinationed of high and low is proposed with (1) and (2).Wavelet detection can inhibit the brink of the most noise in the image, but some of the details are losing in the process of noise suppression due to the brink of the true image often mixed with many noise; Classic Canny operator can get relatively complete information making use of gradient, but these edges are usually drowned by noise if the threshold is selected misconductly, this problerm brings gread trouble to follow-up work, for example, image segmentation, target identification.Canny and wavelet transform are combined in some literatures, wavelet transform is used in the high-frequency part of the image to exact edgess and Canny is used in the low-frequency part, at last the two parts are combined in some fusion rules. The effect of this algorithm is due to the choice of wavelet transform in the high-frequency part and the different fusion rules.Improved B-spline wavelet edge detector in (1) is used in high-frequency part for high-resolution images and atapt space conficcient Gauss wavelet edge detector is used in high-frequency part for low-resolution images. Canny edge detector is used in low-frequency part of all images. At last the two parts are combined in some fusion rules to obtain the last image. Simulation results indicate that comparing with traditional algorithms and B-spline wavelet it is more efficient with both precise image detection and effective noise restraining. However, due to the increase in computational complexity at the expense, time of computus is wasted.