Research On Cigarettes Recognition And Counting Based On Cluster Analysis

Being a method of data processing, cluster analysis has been widely researched in recent years, and has a broad application in image processing, pattern recognition and data mining. The most common method of clustering analysis for image segmentation is the k-means clustering, which need setting a threshold, and then doesn'get people involved. It is very significant for the automatization of image segmentation.For the cigarettes color feature obtaining, first, the paper discusses the selection of color model, and worked out a color feature model of cigarette, which is based on YUV model and simpler than RGB model. Based on the k-means clustering, all pixels in the color cigarette image are divided into background class or cigareete class under YUV model.There are two steps of color cigarette image segmentation. The first step is rough segmenting, according to the clustering attribute of pixel, color of pixel that belongs to the background class is changed into white, the rest that belong to cigarette class keep its original color. The second step is fine segmenting, based on the principle of"nearest in position and least in difference of characters", discrete region is merged into cigarette region.The number of pixels in the circumscribed square of cigarette's cross section is regarded as invariant, the number of pixels that belong to the cigarette class in the circumscribed square is regarded as variable. The threshold is the ratio of the variable to the invariant, based on the threshold, we judge whether a connected region is a cigarette.In order to evaluate the veracity of segmentation, the counting results obtained by the algorithm are compared with the real counting results of cigarettes.