Detection On Prints Color Difference Based On Color Histogram Features

As the second largest economy in the world,the output value of printing and packaging industry in China has been growing steadily in recent years.Printing and packaging of commodities has become an important link between commodity production and consumers in the process of circulation and consumption.Because of the factors such as printing process design,quality of substrate,printing equipment and technical level of operators,all kinds of printing defects will inevitably appear in the printed matter,which will seriously affect the product quality.Not only it makes the enterprise suffer economic losses,but also causes great waste of human and material resources.Therefore,it is an important way to guarantee the quality of printing products to carry out the research on the fast and accurate detection methods of printing defects in the printing production line.Based on the feature extraction of color histogram,this paper proposes an image color detection method which has a certain theoretical reference value for improving the printing quality and reducing the printing cost,aiming at the problem that the efficiency and accuracy of on-line detection of printing color defects are difficult to be balanced.The research work of this paper mainly includes the following aspects:1.For the color standard image "Lena" with an image size of 512 × 512,RGB color space is converted from to HSV color space,and the values of H,S and V components are randomly changed.Then the color difference value between the sample image and the standard image is calculated by CIE L*a*b*,with using 6NBS as the threshold value to judge whether there is color defect in the image.65000 sample images are obtained in the process.According to the statistical distribution characteristics of the image color histogram,the histogram features of the image are extracted by the non-equally quantized method,and finally the numbers of features representing the image color are reduced from 786432 to 784.2.For the problem of color defect detection in this paper,the detection accuracy is taken as the verification index,and the penalty coefficient and hyper parameter of the Gaussian kernel function of support vector machine are 4.5 and 0.1 respectively,which are determined by grid search method.Then,the BP neural network model is constructed before the relationship between detection accuracy and structure of BP neural network model,number of neurons in hidden layer,activation function of each layer,training function time are analyzed.Besides,after the relationship between the convolutional neural network pooling method,the number of convolutional layers,the RMSE of the validation set and the training set,and the accuracy of detection are investigated,the convolution neural network model for the image color defect in this paper is finally determined.3.In this paper,the 10000 sample images of testing set are tested.The results of the experiment show that the detection accuracy of SVM,BP neural networks and CNN are 99.52%,99.7% and 99.77% respectively.And the detection time of a single image is 59.93 ms?65.95 ms and 57.66 ms respectively.The detection accuracy of the three methods is closely related to the number of training samples.With the increment of the number of training samples,the detection accuracy is improved.4.The SVM,BP neural network,CNN and other methods based on the characteristics of color histogram image color defect detection method have significantly improved the detection efficiency,compared with the calculation formula of color difference and super-pixel method.The detection efficiency of CNN method is increased by 3.36 times and 6.11 times,compared with the calculation formulas of color difference of CIE L*a*b* and CIEDE2000,respectively.That means the CNN method can be utilized to detect image color defects through the extraction of non-equally quantitative color histogram features.Not only the high detection accuracy is ensured,but also the detection efficiency is significantly improved.So,this method of CNN shows a good performance in the field of online color quality monitoring of printed products.