Research Of Relative Algorithms For Vision Inspection Of Color Printing

China is a big country of a press, the amount of production of annual printed matter is very huge. The traditional manual work has great subjectivity when measuring printing quality, easily influenced by individual factors. In recent years, there is an increasing need for quality control in modern industry manufacturing. The high cost, low accuracy and very slow performance of human visual inspection has enforced the development of on-line machine vision-based system capable of executing efficient and effective inspection task. The system will help the printing industry to improve the consistency of product quality and lower production costs.However the correct and real-time detection of printing defect is a big problem that requires complex, time-consuming algorithms. The problem of color printing, particularly, is very more obvious. Because of the complication of color representation in color image, the grey image processing technologies can't be directly applied in color image. Without many robust color image processing technologies, color printing is not able to perform real-time inline surface defect inspection. Some attempts have been made to smooth away these difficulties in this essay. The content mostly includes the following parts:Human color vision characteristics are summarized at first. Based on that, various main color spaces, which have been presented by now, are classified and analyzed. And then the transform relationships among those spaces are discussed.The structure, constitution and inspection flow of the inspection systems are designed respectively to fully meet the requirement of the experiment, and also discuss the design basis of the whole framework.Impulse noise can severely affect subsequent image processing; therefore impulse noise filtering, which should have noise-smoothing and detail-preserving qualities, is an essential part of color printing vision inspection algorithms. An adaptive hybrid filter combining a group of sigma vector median filters with different thresholds with a filter based on neuro-fuzzy system is proposed for color image processing.Corner feature based image registration algorithm is selected to realize exact registration of the reference image and defect image. Multivariate mathematical morphology is increasing becoming a powerful utility for color image processing and analysis. A new vector ordering scheme based on distances in the HSI color space is proposed to rank color samples. And an approach based on minimal spanning tree can facilitate to obtain the distance between the supermum and the infimum of color samples within a specified window.Image segmentation is very important for image processing. Because of the nonlinearity of color representation and human vision characteristics, color image segmentation is more difficult than grey image segmentation. According to a proposed dynamic threshold condition, difference image is finally transformed into true defect image after registration based on features with reference image and detected image.A kind of RLE is defined to label defect, various characteristics of shape-defect are extracted to input RBF neural network, which are designed and trained to solve the classification problem of shape-defect kind. In addition, the classification based on the color difference with the regions of the reference image and defect images is introduced to identify kinds of color-defect. The Experimental results demonstrate the validity of those approaches.