一种高效的油液磨粒图像自适应分割方法

为了提高油液磨粒图像的分割效果，优化磨粒自动识别工作的重要环节，提出了一种结合分水岭算法及区域相似度合并的油液磨粒图像自适应分割方法。对于待分割图像，首先通过形态学重构和基于形态学的自适应H-minima技术对梯度图像进行修正，利用分水岭算法完成磨粒图像的一次分割；其次提取分水岭分割后同质区域的Lab颜色特征、局部二值模式（LBP）纹理特征作为区域的量化指标，基于Bhattacharyya系数分别计算区域间的颜色、纹理相似度，设计可以实现权重自适应调整的颜色、纹理特征融合规则，以此来获取同质区域的综合相似度矩阵，实现过分割区域的合并；最后基于统一的后处理算法完成磨粒图像的完整分割。选择60幅磨粒图像对所提方法的分割效果进行测试，结果表明，单幅图像的平均分割速度在12 s左右，分割正确率稳定在90%以上，所提方法避免了用户在分割过程中对图像的交互式处理，较好地平衡了分割效率和分割精度，自适应程度明显提高。 

An efficient method for adaptive segmentation of oil wear debris image

In order to improve the segmentation effect of oil wear debris image and optimize the main content of automatic recognition of wear debris, an adaptive segmentation method of oil wear debris image which combines watershed algorithm and regional similarity has been proposed. First, the gradient image was modified by morphological reconstruction and H-minima technology, and the watershed algorithm was then used to segment the image. Second, after watershed, the Lab color feature and local binary patterns (LBP) texture feature of the homogenous region were extracted as their quantitative indicators, and the color similarity and texture similarity between the regions were calculated based on the Bhattacharyya coefficients. In order to merge the over-segmentation region with much accuracy, an efficient feature fusion rule was designed considering the dynamic weight of color and texture factors. Finally, some post-processing methods were taken to complete the segmentation. Sixty images were selected to test the segmentation effect of the proposed method. The results indicate that the average segmentation speed of single image is about 12 seconds, and the segmentation accuracy is more than 90%. This method avoids the interactive processing when segmenting wear debris images, well balances the segmentation efficiency and segmentation accuracy, and significantly improves the adaptation degree of segmentation program.