Research On Cigarettes Counting Based On Tobacco Color Model And Template Matching

Image Recognition for cigarettes counting is a new field of image processing application. Compared to mechanical or manual way of counting, it could use less time without cigarettes damage. Generally binarization is a first step in image processing, which can simplify the information, and quickly extract the features from the image. Meanwhile it creates a new possible problem: The recognition accuracy will also depend on the quality of this binary image. To address the problem, recognition directly based on the color image is a preferred way, which uses the characters of tobacco color distribution and the cigarette's shape.Real-time access to accurate, stable cigarette image is the basis of image recognition. The appropriate light source selection and the location of the equipment is the key issue in the image acquisition system. And a realization of the image acquisition system is given with the combination of actual production demand.Cigarette image preprocessing's main target is to remove impurities outside the cigarette region, and extract features of the objects. It can be divided into two steps. First, use the color segmentation method to remove the background impurities. Secondly, use the boundary detection method to find the frame of cigarette container, which assumes that the boundaries of the cigarette container are linear, and then remove all the regions which outside the boundary.Through the Tobacco color experiments analysis, the tobacco color distribution's features are as follows: a) The tobacco color distribution is concentrated while brightness changes in a certain area; b) Hue of the Tobacco color rare changes when brightness changes; c) Saturation changes in a nonlinear parabola way when brightness changes. Based on these features, tobacco color model in HSI and RGB color spaces is constructed, and make the segmentation of cigarette internal and external region through the tobacco color model.The cigarette circular template is constructed before cigarettes Counting. And the template is divided into six sub-regions for the facilitation of experimental observation. Observe the distribution of tobacco pixels and non-tobacco pixels, and then judge a circular region as a cigarette compared to the sample library.The arithmetic is analyzed and compared from the performance in the field of recognition accuracy and the system's stability.