摘要
基于轮毂缺陷目标的尺度特征,利用纵横两个方向的"山峰"定位,获得轮毂铸件缺陷像素小块,并提取出缺陷目标区域;对目标区域进一步提取种子点,建立生长障碍,对种子进化处理以及图像二值化,利用改进的种子填充法将缺陷用二值图像完整清晰地表达出来,实现基于缺陷目标特征与种子填充法的轮毂铸件缺陷检测方法.以实际检测中4种常见的不同类型缺陷、不同结构部位、不同背景等的轮毂缺陷图像作为实例,对100幅缺陷图像进行算法验证.实验结果表明,该算法不需要对图像进行预处理,可以检测各种缺陷类型,对对比度低、高噪声、产品结构和图像背景复杂、光照不均匀的图像也能进行处理,不需要标准参照图片,运算速度满足工业应用的要求.
The method presented was based on the characteristics of the size of wheel flaws, using horizontal and vertical hill-locating to get blocks of flaw pels, and to pick up the goal flaw-located areas. The goal flaw-located image was used to get the seeds, growing obstacles, seed-evolution and initial casting image thresholding. The binary-valued image obtained by the improved seeded region growing method displayed the flaws of the casting image accurately. Four samples, i. e. , four images with different flaw types, different parts of casting and different backgrounds, were used to present the method. And it was tested by 100 images. The result denominated many advantages of the method. It did not need complicated pretreatment, but could deal with diverse types of flaws. Images with poor quality （i. e. , poor-contrast, high noise, complicated structures and backgrounds, asymmetrical illumination and ete） could be processed, and reference images were not needed. The processing speed met the demand of industry uses.
出处
《浙江大学学报（工学版）》
EI
CAS
CSCD
北大核心
2009年第7期1230-1237,共8页
Journal of Zhejiang University：Engineering Science
基金
国家自然科学基金资助项目(50775201)
浙江省自然科学基金资助项目(Y107431)
