摘要
针对芯片X光图像多目标、背景复杂、灰度分布不均匀的特点,提出一种基于Mean Shift的层次分割算法。运用旋转缩放不变的模板匹配定位算法定位芯片并确定其兴趣区域(ROI),从而实现外层分割。再运用Mean Shift算法对芯片模板图像和芯片ROI图像进行统计聚类分析,分别计算原图像金线模式类的灰度平均值,以芯片模板图像的金线模式类的平均灰度为基准,对芯片ROI的聚类图像采用优化阈值进行自适应阈值分割,从而实现内层分割。实验结果表明:与传统的均值偏移分割算法相比,该方法的多目标分割准确,可靠性高。
To overcome segmentation difficulties induced by multiple targets of interests, redundant background and inhomogeneous gray levels of the X-ray image, a hierarchical segmentation algorithm based on Mean Shift is proposed. The algorithm is achieved through two layer segmentation. The outer layer segmentation is to locate the positions and determine their regions of interests (ROI) for all single chips in the X-ray by the matching algorithm with the template of scale invariance and rotation invariance. The inner layer segmentation is to extract sub- objects, i. e. , gold lines, balls and PAD of each chip by clustering and analyzing the ROI image and template image using the Mean Shift algorithm. Then calculates mean gray values of gold lines-pattern regions for ROI and template images respectively. Finally, according to the mean gray values of sub-patterns in the template, design an optimized threshold and segment the ROI image by the adaptive threshold segmentation algorithm. The experimental results show that the proposed segmentation algorithm achieves more accurate and reliable results compared with traditional algorithms.
出处
《传感器与微系统》
CSCD
2016年第6期128-131,共4页
Transducer and Microsystem Technologies
基金
国家自然科学基金资助项目(61305016)
江南大学自主科研计划青年基金资助项目(JUSRP1059)
关键词
Mean
SHIFT
层次分割
兴趣区域
自适应阈值
Mean Shift
hierarchical segmentation
region of interest(ROI)
adaptive threshold