Research On Color Grading And Defect Detection Of The Tiles Based On Computer Vision

In this paper, the research is mainly focused on the theory and algorithms of floor& wall tiles'surface quality inspection based on computer vision, including defectdetection and color grading. The following research results and conclusions areobtained after lots of work conducted on reference reviewing, investigation on field,analysis and development of algorithms and experimental sduty:1. Firstly, an overall review on the development and application of visualinspection technology is presented, and also introduced is their application to surfacequality inspection and control. Otherwise, some subjects involved with computervisual inspection are discussed briefly.2. The Current research methods on surface defect detection and color grading offloor & wall tiles are discussed in detail, and it is pointed out that therir applicationunder some situation are limited because of not taking texture and scale into account.A new point of view is purposed to utilize advanced texture analysis technology.3. Image acquisition and pre-processing of tile is studied, and three illuminatingdesigns are proposed and compared. Also studied is calibration of camera, denoise andsegment of image.4. To detect spot-like defect on ceramic-glass surface, which couldn't be dealtwith classical methods, Markov Random Field texture analysis is adopted. The texturecharacteristic has been strengthened before analysis. MRF has good ability to describestochastic texture. Experimental result has shown the new method performance wellwhen used to detect irregular spot-like defect.5. In current research and report, the grading performance is bad when classifythose tiles with stochastic texture and multi-color surface. To improve that, thecombination of color, texture and scale is purposed in this paper. The combinationcharacteristic set is acquired by wavelet analysis of color image. In addition, anoptimization algorithm has been operated to get sub-optimal characteristic set. It'sverified feasibility and effective to solve the difficulty of color grading of tiles withstochastic texture and multi-color surface.6. The application of ART2 net served as a classifier to ceramic-glass tiles isstudied, and modified learning algorithm is proposed to improve the pattern shiftingproblem existed in classical ART2 net. Good performance of color grading has beengained when the classifier was used to ceramic-glass tiles without learning samples.