基于增强的低对比度印品缺陷的识别技术研究

目的 针对当前印刷缺陷检测系统中存在的低对比度印刷缺陷检测精度不高等问题，基于HSV颜色空间，提出一种增强的低对比度印刷缺陷识别方法。方法 首先，将标准样张图像与采集到的印刷图像由RGB颜色空间转换到HSV颜色空间，并提取视觉上变化敏感的亮度分量V作为待检测对象；其次，将对比度受限的局部直方图均衡（CLAHE）与数学形态学相结合，来增强显现待检测图像中的缺陷；再次，使用连通域分析方法来获取缺陷的面积、周长、离心率、长宽比和圆形度等5种特征信息，并以此建立15个特征模型；最后，构建基于PNN的印刷缺陷识别神经网络，并在Matlab中实现对低对比度印刷缺陷的识别。结果 15个模型的平均耗时为475 ms，都控制在毫秒级别，满足了现代印刷缺陷检测对于实时性的要求。其中模型2的测试正确率为95%，能够识别污点等点缺陷，模型3和模型12的测试正确率为93%和93.3%，能够识别刮痕等线缺陷，模型5的测试正确率为93.1%，能够识别墨迹等面缺陷，且测试正确率高于基于BP神经网络的缺陷识别方法。结论 从缺陷检测的实时性和精确性上来讲，提出的方法能够对低对比度印刷缺陷进行实时和精确的检测。

Recognition Technology of Low Contrast Printing Defects Based on Enhancement

The paper aims to propose an enhanced recognition method for low contrast defects based on the HSV color space to solve the problem that the low detection accuracy of low contrast printing defects in the current printing defect detection system. Firstly, the standard sample image and the printed image to be detected were converted from the RGB color space to the HSV color space, and the brightness component V, which was sensitive to visual changes, was extracted as the object to be detected; secondly, the Contrast Limited Adaptive Histogram Equalization (CLAHE) was combined with mathematical morphology to enhance the appearance of defects in the image to be detected; thirdly, the Connected Component Analysis (CCA) was used to obtain the five kinds of characteristic information of area, circumference, eccentricity, length-width ratio and circularity to establish 15 feature models on this basis. Finally, a printing defects recognition network based on PNN was constructed, and the recognition of low contrast printing defects was realized in Matlab. The average time of the 15 models was 475 ms, all of which were controlled at the millisecond level, meeting the real-time requirements of modern printing defect detection. Among them, the test accuracy of model 2 was 95%, which can identify spot defects; the test accuracy of Model 3 and Model 12 was 93% and 93.3% respectively, which can identify line defects; the test accuracy of model 5 was 93.1%, and it can identify surface defects. Moreover, the test accuracy was higher than that of the defect recognition method based on BP neural network. In terms of real-time and accuracy of defect detection, this method can detect low contrast printing defects in real time and accurately.