基于第2代Curvelet的非监督式纹理缺陷分割

Unsupervised defects segmentation of texture based on second-generation curvelet

针对纹理缺陷分割问题,将曲波变换与均值漂移理论相结合,形成有效的纹理分割新方法。首先,通过曲波变换将图像分解到各通道,对各通道的图像进行非线性变换得到特征图像;然后,用均值漂移算法对各通道特征图像进行自适应聚类,找到各通道的奇异点;最后,对所有通道滤波后的图像进行重构,使缺陷凸显并通过阈值法二值化。该方法不需要学习样本,可以快速、精确地定位到多目标物边界,对旋转、亮度变化、噪声、弱边界具有很强的鲁棒性。通过MATLAB进行仿真实验,验证了该方法的有效性。

For the problem of texture defects segmentation, a new texture image segmentation approach based on the Mean Shift theory combined with the Curvelet transform is proposed. First, a Curvelet transform is used to decompose the image to each channel. Secondly, each channel＇s feature image derived from non-linear transformation is adaptively clustered to find the singular points using Mean Shift. Finally, the filtered images of all channels are reconstructed to make defects prominent, and the binary image is obtained by a threshold. In this paper, a learning sample is not needed and the multiobjects boundary is located fast and accurately. This method is robust against rotation, brightness changing, noise, and weak boundaries. The effectiveness of the method is verified by MATLAB simulation experiments.