【摘要】 铝合金低倍组织图像存在缺陷分布稀疏、面积极小等检测难题，一般的图像分割法难以将缺陷从背景中分割出来。为此，设计一种基于局部分类的阈值分割法。对图像进行去噪及增强处理以突出缺陷特征，使用滑窗法对图像进行局部阈值分割，由分割结果的轮廓特征划分缺陷区域与无缺陷区域，再根据其各自的阈值分布确定全局阈值，计算图像的错误分类误差以验证算法的有效性。研究结果表明：所提出的方法相比Otsu法和最大熵阈值分割法可更有效地分割铝合金低倍组织图像，适合铝合金低倍组织的缺陷检测。更多还原

【Abstract】 To overcome the difficulty in segmenting the defects from the background by general image segmentation method due to poor detections of aluminum alloy macrostructure image such as sparse defect distribution and small area, a threshold segmentation method based on local classification is proposed. The image is denoised and enhanced to highlight the defect features. The sliding window method is used to segment the image by local threshold, and by the contour features of the segmentation results, the defect area and the defect free area are divided. The global threshold is determined according to their respective threshold distributions, and the error aviation of classification error of the image is calculated to verify the effectiveness of the algorithm. The experimental results show that the proposed method is more effective than Otsu method and maximum entropy threshold segmentation method in segmenting the image of aluminum alloy macrostructure, which is suitable for defect detection of aluminum alloy macrostructure.更多还原