基于LS-SVM的contourletHMT变换的孔型测量图像边缘检测

Edge Detection for Drawing Die Measure Based on Least Squares Support Vector Machine in a Contourlet HMT Domain

为了提取拉丝模具内孔的衍射图像的边缘信息,以往的检测方法都有各自的局限性,因此采用contourle tHMT变换结合LS—SVM的方法;首先利用contourlet HMT变换将图像分成粗糙系数和精细系数,通过LS—SVM训练粗糙系数找出边缘系数和噪声系数;利用BayesShrink方法去除噪声系数,采用梯度边缘系数和去噪后的系数,用加权平均值的方法进行图像的融合,最后使用contour—let HMT的逆变换对图像进行重构;对不同情况下的图像进行了测试,结果证明Contourlet HMT的边缘检测方法明显优于其它的方法。

In order to extract the edge information of the diffraction image of the inner hole in the wire drawing die, with its own limitations of conventional detection method, so the contourlet HMT transform is used combining LS--SVM. Firstly contourlet HMT divides an image into coarser coefficient and finer coefficient. Train the coarser coefficients and classify them using the LS--SVM classifier into edge and noise coefficient classes. Remove noise coefficient using the BayesShrink method and obtain a denoised coefficient class. Detect edge coeffi- cient class and denoised coefficient class using gradient. Fuse edges information using weighted average fusion rule. Finally, reconstruct edge image using inverse contourlet HMT. The image is tested in different conditions. The results show that the edge detec- tion method of HMT contourlet is better than other methods