Marble Class Identification And Application Based On Improved K-means Algorithm Classification

Marble grade identification is a key link to achieve full automatic processing in marble enterprises and also an important problem of content based classification in image processing.Although studies on image classification have made great progress,the related algorithmic research is slow,due to the complexity of color and texture features.Based on this consideration,this dissertation,based on the problems of feature extraction,image classification and algorithm-based quality identification,not only probes into the models of feature extraction on color and texture,but also designs both an improved K-means algorithm suitable for image classification and a simple and available marble grade identification system.It is helpful for the later development of the research of marble classification and also of the important reference value for identifying the grade of marble quality.The main work and achievements acquired can be summarized below.1.In view of the problem of color feature extraction of marble images,the statistics are done with the aspect of image feature information by means of histogram formulation after linear combination on shade,tint,and brightness,which depends on the relation between RGB and HSV color spaces.Further,a color feature vector model is acquired by improving the reported computational model on inertia ratio and the pyramid model.Numerical experiments have showed that the feature vector model can reflect marble images' color features.2.In order to address the problem of texture feature extraction of a marble image,a primitive matrix model is developed by constructing several image texture units and designing a primitive matrix model.This,along with the reported LBP operator and four feature vectors of energy,inertia,inverse moment and distribution uniformity,derives a texture feature model to measure the texture feature of the image.Numerical experiments have illustrated that one such model can effectively identify different kinds of texture features.3.In order to avoid the influence of initial center points to classification quality and the difficulty of deciding the number of classification for the original K-means algorithm,an improved K-means algorithm for content-based image classification is proposed.Additionally,a simple software system is developed to solve the problem of marble grade identification,based on MFC platform.As integrated the feature vector models of color and texture with the improvedK-means algorithm,the system is acquired after eliminating the background information of marble images by improving the threshold selection scheme of the method of maximum inter-class variance.Numerical experiments have validated that the software system can not only effectively reflect the feature information of the images,but also identify the grades of different kinds of marbles.