【作者】 王世允；

【导师】 刘世昌；

【作者基本信息】 西安理工大学 ， 制浆造纸工程， 2002， 硕士

【摘要】 为了克服控制条在印品质量检测中应用的弊端，本文提出了对印品图像直接检测的思路。用数字图像处理的方法对被测图像进行严格定位，解决了前人机械定位不准的问题。并用模式识别方法分割网点，继而得到网点面积率，解决了纽介堡方程在图像直接检测中不能计算网点面积率的问题。本课题的研究为印品图像质量直接检测技术扫除了根本的技术关键。 在定位技术中，先将标准图像与待测图像旋转，使图像坐标系与平面直角坐标系之间不存在角度，然后从待测图像的特征图像中抽取一定大小的子图像A，让它在标准图像的特征图像中逐行逐列扫描，找到与该子图像相似度最大的子图像B，记下B的坐标，利用几何坐标平移和旋转的知识进行定位。这能使两幅图像的坐标位置最接近，图像定位的准确度提高到象素量级。这样才使后来的检测更具有说服力。 在定位准确的基础上，利用模式识别技术将指定位置、指定大小的子图像上各色网点识别出来，才能对网点的参数进行分析，完成印刷图像质量直接检测。本课题网点的识别是以其颜色为特征向量的。通过Lab距离判决法和神经网络识别模型，比较识别后计算的网点面积率与纽介堡方程计算的网点面积率，证明模式识别方法可以用来分割网点。并从理论和实践上分析比较，证明神经网络建立的识别模型更适于网点的识别。更多还原

【Abstract】 To overcome the defect of control strip in printing quality testing, the idea about printed image testing directly has been suggested in this article. The being tested images have been located strictly by the means of digital image processing to make the location accuracy better to by mechanical means. Because Neugebauer equation can’ t compute dot coverage area in general image testing directly, the dots in primary color have been segmented by pattern recognition, and the area of every colored dots have been computed secondly in this article.In the chapter of locating, the standard image and the being tested images have been rotated firstly to remove the angle between the image coordinate and the right-angled coordinate in a plane, a sub-image A with a fixed area has been cut at the fixed location in the being tested image and moved in the standard image each line and each volume to find the most similar sub-image B. The location of sub-image B has been recorded, the most similar two pairs of location has been acquired by the means of Geometric moving androtating.On the base of locating, the different colored dots have been classifiedby pattern recognition in the fixed location and area sub-image to analyzethe parameter of dots and finish image testing directly. The colors of dothave been made to be characterization vector in order to train recognition model in this article, classifying models have been set up by decision function of distance and neural network, the data about dot area coverage from classifying models have been compared with that from Neugebauer equation to prove that different colored dots can be classified by pattern recognition. And the data has been analyzed in theory and practice to prove that the neural network model is better than statistic model to classify the printed dots.更多还原