Research On Ceramic Tile Image Classification Method Based On Decision Tree

As the ascension of market demands for quality of ceramic tile product and efficiency of the production, the current manual sorting approaches have been used cannot meet the production needs. Enterprises urgently need to develop machine vision based ceramic tile classification and sorting system to solve problems such as high cost of production and non-uniform of classification standards. This paper mainly focused on algorithms of how to quantize the characteristics surface of ceramic tile and classify them.Firstly, a correction method based on mapping was utilized to resolve the uneven illumination and geometric distortion. By analyzing dominant color of the rectified images, an improved color quantization method based on HSV color model was designed. The outputs of main color method combined with the image average gray level were all as the color feature parameters. On the texture analysis, the algorithm based on Radon transform was used to extract the texture direction and a statistical method was adopted to get the texture distribution ratio.Secondly, on classification algorithms, the number of training samples is limited in the initial stage of acquisition and classification features require to be filtered adaptively. To solve these problems and meet the demands of accuracy and real-time performance, the decision tree algorithm based on feature distances and statistical distribution was designed. The construction algorithms of decision tree under the condition of both single class sample and a large number of samples was elaborated. Then, how to determine feature weights and feature association rules were discussed. The algorithm improves the segmentation criteria of the characteristics. It can obtain the split positions between property values more accurately, and improve the accuracy of the classification prediction.Finally, the same training and test sets were used to conduct experiments among the decision tree, minimum distance classifier, support vector machine, C4.5and CART algorithms. Comparative analysis of the experimental results proves the decision tree algorithm has advantages in classification accuracy and real-time performance.