一种改进的高斯混合模型的玻璃缺陷分割方法

针对传统玻璃缺陷检测技术准确率较低、时间长、精度低等难点,提出了一种改进高斯混合模型的玻璃缺陷图像分割方法。首先,基于分数阶微分运算获取灰度特征,并利用灰度共生矩阵提取纹理特征,构建玻璃缺陷完整的双特征观测数据;然后,引入相邻像素间的空间关联性和约束性,通过交替进行基于双特征随机场评估像素点与标号场之间的对应关系和空间约束来完成玻璃缺陷分割;最后,在不同温度系数参数β下对分割算法进行了性能测试实验,同时,与当前流行的分割算法对4种不同类型的玻璃缺陷进行了性能比较实验。实验表明该算法能够提高图像分割的鲁棒性和精确性。

Improved Gaussian Mixture Model for Glass Defect Image Segmentation

Aiming at the difficulties of low accuracy,long time and low precision of traditional glass defect detection technology,an improved glass defect image segmentation method based on Gaussian mixture model(GMM)was proposed.First,gray-scale features were obtained based on fractional differential operations,and texture features were extracted using a gray-scale co-occurrence matrix to construct complete dual-feature observation data for glass defects.Then,the spatial correlation and constraint between adjacent pixels were introduced,and the glass defect segmentation was completed by alternately evaluating the correspondence between pixels and labeled fields based on the dual feature random field and the spatial constraints.Finally,the performance test of the segmentation algorithm was carried out under different temperature coefficient parameterβ.At the same time,the performance comparison experiments of four different types of glass defects were performed with the current popular segmentation algorithms.Experimental results show that the proposed algorithm can improve the robustness and accuracy of image segmentation.