Recognition Research Of Ceramic Tiles Classification Based On Machine Vision

China has a lot of the production of ceramic tile company.In many small andmedium-sized companies due to production conditions, cost and other factors, avariety of pattern tiles share a production line. In this company, not only the highlabor intensity, while inefficient, error rate. There are some tile-related research, forexample, the tile color classification, and surface inspection, but none focus ondifferent categories, classification and identification of different patterns of tiles.Wide production line required multiple cameras to collect tile pictures, it should firstget to the picture mosaic, and then only enter the identification step.The purpose ofthis study is image mosaic,the tile classification and methods of feature selection andclassification is the main part of the study.This paper first outlines the various methods of image mosaic, and pointed outthat the method based on feature points is adopted. The principle of Harris and SIFTwas detailed analyzed, and gives the detection results. How the RANSAC algorithmworks and the implementation steps of solving the transformation matrix H werediscussed. Finally, the the mosaic ceramic pictures spliced by SIFT feature pointswere shown.Tile feature is extracted from three aspects of color and the edge moments. Avariety of color space and the conversion between them are discussed in. Therepresentation and quantization of color in HSV space. By the formula, thethree-dimensional HSV color space is compressed into one-dimensional space. Colorvalues in the one-dimensional space is limited within the range of values of0-71,atotal of72, which guarantees efficient and simple follow-up of the color featureextraction. after compression and quantization, the color histogram is calculated andthen Extract the mean and variance of the then three major color components. Manyedge detection algorithm is discussed, and then select morphological methods todetect tile edge. After edge detection,two invariant moment the edge of the wereextracted from each block. The resulting characteristics of the location also given adetailed exposition to give corresponding solution. Feature dimension is reduced by PCA.Using the classification and identification of the BP neural network to obtain thefeature vector tiles. The test results show that the proposed method has higher correctrecognition rate. The BP neural network is employed to classify different ceramic tile.The test results show that the proposed method has a higher correct recognition rate.