基于反对称双正交小波变换的多尺度归一化分割方法-计算机辅助设计.PDFVIP

第28 卷 第 1 期 计算机辅助设计与图形学学报 Vol. 28 No.1 2016 年 1 月 Journal of Computer-Aided Design Computer Graphics Jan. 2016 基于反对称双正交小波变换的多尺度归一化分割方法 * 王 森, 伍 星, 刘 韬, 张印辉 (昆明理工大学机电工程学院 昆明 650500) (xwu@kmust.edu.cn) : 针对工业检测现场拍摄的微电子元件图像产生的模糊和噪声叠加现象, 提出基于反对称双正交小波变换的 多尺度归一化分割方法. 首先通过反对称双正交小波的多尺度边缘检测方法对待处理图像进行平滑、去噪及轮廓提 取；然后利用重构后的边缘轮廓、强度值和图像的约束矩阵求得权重矩阵; 最后利用谱分割技术得出图像的特征向 量, 并离散化后得到最终的分割结果。对工业显微镜采集的电路板零件和部分故障图像以及PASCAL VOC2012 分割 数据集中的图像进行测试, 并与归一化割方法、多尺度归一化割方法、最小割/最大流方法和基于约束参数的最小化 割方法在精确率、查全率、F-测量、平均绝对误差和运行时间上进行了对比, 结果表明, 该算法的分割质量更理想. ：图像分割; 反对称双正交小波变换; 多尺度归一化割; 反对称双正交小波变换的多尺度归一化割 ：TP391.41 Multi-Scale Normalizaed Cut Based on Anti-Symmetrical Bi-Orthogonal Wavelet Transform * Wang Sen, Wu Xing , Liu Tao, and Zhang Yinhui (College of Mechanical and Electrical Engineering, Kunming University of Science and Technology, Kunming 650500) Abstract: A multi-scale normalized cut algorithm bases on anti-symmetrical bi-orthogonal wavelet transform is proposed to address the problems that blur and noise occurred in the image of Micro-electronic components for industrial inspection site. Firstly, multi-scale edge detection of the anti-symmetrical bi-orthogonal wavelet is used to extract the contours with smoothing and denoising. Then, according to constraint matrix. edge contours and strength values can obtain a weight matrix after decomposition of the wavelet of image. Finally, the eigenvectors of the image are extracted easily using the spectral segmentation techniques, and the segmentation result is ob- tained after discretization. To test the effectiveness of the introduced scheme, the image segmentation tests are carried out for t