Study Of Automatic Inspection Technology On Quality Of Tiles

Based on technique of image processing, pattern recognition and artificial intelligence, theories and methods of automatic inspection on quality of tiles are studied in this dissertation. Some creative results are achieved as follows:1. After analyzing the usual filter de-noising methods of grey images and color images, and the tile's characteristics, a new fast median filtering method based on the sparse distributing windows is presented. The sparse distributing windows fitting to different variety of tiles are presented by experiments. This method can saves memory rooms, improves efficiency and operation rate.2. The principles and methods of Canny algorithm and Radon transform are discussed in detail. The threshold value selecting of Canny algorithm fitted to the tile edge extraction is proposed, furthermore a matching method based on tile's edge character is presented. In this method, the tile's edge feature is extracted by Canny operator and Radon transform first, and then the matching is made by rotating and shifting the image. The application of this method to surface defect detection of tiles is introduced.3. A new image matching method is presented in this paper. This method is an improvement of cross-correlation method that can be used in regular pattern tiles. The experiment shows that this method can get more precise matching after the primary matching based on tile's edge matching method.4. After discussing the automatic visual inspecting method based on co-occurrence differential Stat., two new methods are presented and applied to tile's automatic defect detection. One is based on image division and local hue accumulating histogram. Feature vector with color space and stat. characteristics can be got by this method; the other can construct a criterion vector combining with co-occurrence matrix texture features and color features. Both are feasible in automatic quality inspection of tiles.5. A new feature algorithm is proposed based on wavelet transform in HSI space to obtain a feature vector with combining information of color, texture and scale. Both the new algorithm and the BP neutral network are used in automatic quality detection of those tiles with stochastic texture and...