中文摘要：

现有的SIOX等主流图像分割算法对印刷电路板（PCB）CT分层图像中的过孔分割质量较差,而实际中常用的基于Hough变换的分割算法在PCB图像噪声大时容易出现分割错误。为解决以上问题,提出一种基于超像素的PCB CT图像过孔分割算法。首先通过对比实验在六种常用的超像素分割算法中选择一种最适合PCB CT图像的算法进行超像素分割,利用具有局部特征表达能力的超像素代替像素作为分割的基本处理单元。针对过孔表现出的显著圆形特征,引入圆度率对超像素的形状特征进行描述和表达,在此基础上设计一种全新的超像素合并与筛选策略来提取过孔目标,最后通过形态学处理得到最终分割结果。通过实验对比,选择ERS算法对PCB CT分层图像进行超像素分割;与基于Hough变换的分割算法相比,新提出方法在分割准确率和召回率两方面均提高了约10%,特别是在图像背景噪声较大时出现错误分割的概率明显小于基于Hough变换的分割算法。基于超像素的PCB CT图像过孔分割算法克服了PCB CT图像噪声大、图像灰度不均匀等情况对分割带来的困难,能够准确地从PCB CT图像的复杂背景中提取出过孔目标。

英文摘要：

The mainstream image segmentation algorithms like SIOX（ Simple Interactive Object Extraction） hardly extract the via holes from Printed Circuit Board（ PCB） CT monolayer images satisfactorily,and the Hough transform based segmentation algorithm which easily makes errors while the PCB CT monolayer images are of poor quality is commonly used in a real world application. To meet the demand for accurate segmentation of PCB via holes,a segmentation method was proposed by using superpixel. First,superpixel was introduced to take the place of pixel as the processing primitive in segmentation,which has ability of local features representation. The most proper algorithm for superpixel segmentation of PCB CT monolayer images among six popular superpixel segmentation algorithms was chosen through experimental results. Considering the shape characteristics of PCB via holes,circularity was introduced to describe the shape feature of superpixel. Furthermore,a novel superpixel merging and pruning strategy based on circularity was proposed to extract the via holes from background. Finally,the morphological image processing methods were used to improve the final segmentation quality. The experimental results show that the ERS（ Entropy Rate Superpixel） segmentation algorithm is best for PCB CT monolayer images,and the average precision rate and recall rate of the proposed method both are approximately ten percent higher in comparison with the Hough transform based segmentation algorithm. The proposed method can solve the difficulties of much noise and intensity inhomogeneity in PCB CT monolayer images,and extracts the via holes accurately from complex background.