玻璃瓶底污物的实时检验图像识别方法

玻璃瓶底残留污物的实时自动检，其难点在于瓶底图像复杂多变，耍设法把由瓶底本身凸凹结构或标记等形成的暗影与真正污物区分开。针对实际生产的需要，捉出一种玻璃瓶底实时自动检验的计算机图像识别方法。主要包括分区选择多级自适应特正门限，通过区域生长方法提取污物特征参数，判决分类。此方法有效地解决了实时性与准确性要求的矛盾，并已在模拟验瓶系统上用软件实现。实验与测试结果表明，在检测直径为 ２ｍｍ不透明污物的最小溶限下，正确识别率为 ９６．４％，漏检率０．４％，误检率３．２％．能满足实际生产中每秒检验１０个瓶于的需要。

Method of image recognition for real time automatic inspection of glass bottle bottom

A method of computer image recognition for real time automatic inspec- tion of dirt remained on the glass bottle bottom is outlined. The image of each bottle bottom is first divided into three subregions according to the properties of the image itself.Individual subregion threshold and variable threshold of a local neighbourhood of each dirt are determined to detect dirt.Then an algori- thm of the fast region growning is utilized to detect all pixels of the dirt by the subregion threshold and local threshold of neighbourhood of each dirt. The features such as the area,perimeter, maximum length,maximum width and average gray level of each dirt are extracted simultaneously. Finally according to designed criterion, the real dirt is distinguished from the shadows of bottle bottom itself. The complete algorithm has been implemented by 8088 assembly language on a simulated system of which the ratio of successful inspec- tion is more than 95 percent. The time to process an image is less than 90 ms so that it is possible to inspect more than 10 bottles per second required in a real system.