Research On Cigarette Image Segmentation And Counting Based On Genetic Algorithm

Cigarette counting is a necessary working procedure in the manufacture workshop of medium-sized or pint-sized cigarette factory. The use of image processing technology to do cigarette counting could be an optional scheme. Recognition directly based on the color image avoids color information loss in the binary image However, the efficiency and the accuracy of the recognition algorithm is not high. In order to improve or solve the above problem, the genetic algorithm for multi-target recognition was applied to do cigarette counting.Cigarette image segmentation is an important step before cigarette counting. A good segmentation is the prerequisite of cigarette counting. Based on the analysis of cigarette image, we combine Genetic Algorithm and the maximum entropy principle to extract the segmentation threshold of each color channel, the segmented image with the largest entropy based on the threshold was identified as the final result.According to round feature of the cigarette, a circular template was defined in the cigarettes counting section. The part hausdorff distance was introduced to genetic algorithm as the fitness function. The fitness function measures the similarity of cigarettes and the template to determine whether the identified area is a cigarette or not. Using genetic algorithms to identify cigarette and counting, is a multi-object counting. However in the past genetic algorithm is applied only to the single object of matching identification, even for multi-object it only applied to the segmentation stage. Therefore this method is not just for the most optimum answer. Rather than the individuals of whose fitness value is greater than the threshold value in each generation are seen as answers, this would resolve a number of target identification problems.Based on analyzing segmented cigarette image, three different solid round templates were defined. They are combined with genetic algorithm to randomly generate the coordinates of the matched region center, and then the matched region was localized. The comparison between the matched region and the templates was applied to identify and count the cigatetteThe preliminary experiments have analyzed the efficiency and the accuracy of the algorithm, and the method for improving the whole accuracy was also presented.