Research And Design Of On-line Color-difference Evaluation For Dyeing Products Based On Machine Vision

In the dyeing and printing field, textile and dyeing enterprises must strictly control theimportant color quality indicator which can’t be ignored to improve the market competitivenessof the finished piece goods after dyeing. Traditional color difference detection mainly relies onthe artificial to complete. The strong subjectivity and low efficiency exist in the detectionprocess. So it is important and necessary to consider the factor. It is a good solution to introducethe machine vision into the chromatic aberration detection field to replace human visualinspection. This solution has a large advantage. Based on machine vision and image processingtechnology, this paper has studied the color detection theory and algorithm, and applies theresearch theory to the testing evaluation of wide width and large angle product to solve the keytechnical problems such as image acquisition, image matching, image feature extraction andcolor constancy description.Firstly, this paper has studied the technical theories, such as lighting system, camera, imageacquisition card. On this basis, the suitable hardware is chosen for the system. The overallplanning of the "online color difference detection system based on machine vision" has beenschemed. Besides, the dynamic experiment platform is designed and built. The contrastexperiments of the image preprocessing algorithm performance and three kinds of colordifference formulas have been carried on. According to the experimental results, the suitableimage processing algorithms and color difference formula is chosen for the system.Due to the need of collecting images of wide cloth, two sets of color industrial cameras ofhigh performance and high precision accuracy are installed to capture images, and then transferthe collected data to the personal computer(PC) through the PCI gigabit card. At first, the imagehas been pre-processed. After the image registration and image fusion steps, a high resolutionand large field wide surface color images will be joined each other. It can realize fast imagemosaicing and guarantee the quality of stitching images.In the color difference detection of dyeing product, the unstable illumination change willseriously affect the evaluation results. So, the improved new anisotropic diffusion model(Perona-Malik model) is introduced. After the illumination invariant process of traditionalanisotropic diffusion model and the proposed improved anisotropic diffusion model, with Datacolor650color measurement instrument measurements as a standard, image processingeffect is analyzed finally by using the color difference formula CIEL*a*b*which is commonlyused in the textile industry. And the experimental results show that the illumination invariantimages obtained by the improved image denoising algorithm have no the phenomenon of haloand spiny and have better image smoothing effect, are in line with the human visual system andprovide security for the illumination invariant color difference detection of dyeing product.In order to better solve the problem of color constancy, according to the method ofdetermining the error limit on the basis of distance between the standard samples and the testsamples, least squares support vector regression (LSSVR) algorithm was improved, and correctsillumination of different colors of textiles under the irradiation combined with the improvedmodel which refers the average of X(1)(1)~X(1)(n) as the initial value of GM(1,1). Gray theoryhas good local optimization performance, makes up the defect of the LSSVR’s falling into theglobal optimal, improves the textile illumination correction effect better. Contrast experimentresults show that the method has good stability and the correction effect, and LSSVR reduces thetime of illumination correction by adopting the method of fractional processing small data. Theimproved grey prediction model combined with improved improves the prediction accuracy.Finally, in order to quantify the dyeing color difference evaluation results, this paper putsforward the color difference evaluation algorithm of consistency and uniformity and adopts theprincipal component analysis(PCA) to reduce the data dimension to further analyze it. Supportvector machine (SVM) algorithm can well solve the problem of small sample data classification.The color difference evaluation model is established based on the machine learning algorithm,and the Naive Bayesian Model (NBM) and least squares support vector machine (LSSVM)algorithm are used for consistency and uniformity of data modeling. Experimental results showthat compared with the traditional naive bayesian method, the evaluation algorithm andevaluation model put forward by this paper can realize evaluation for dyeing effect of dyedgoods fast and exactly.