Research On Acquisition And Binarization Of Cigarette Images

In a tobacco factory, the specific number of cigarettes produced requires counting. But the counting precision of traditional mechanical counting method is not high enough. Also, the traditional method will damage the cigarettes. In light of this situation, using image recognition methods to count cigarettes is a better choice with the virtue of high speed and high accuracy. Above all, the method does not damage the cigarettes. Using image recognition methods to count cigarettes include three steps: cigarette image acquisition, image binarization and recognition and counting of cigarettes.After comparing all kinds of image acquisition equipments existed, digital camera is selected as the image acquisition equipment used for cigarette counting application. After configuring the digital camera parameters, the images obtained using digital camera to shoot the cigarette containers are pictures of Bayer, which need interpolated for human visual observation. Therefore, in the step of image acquisition, the missing values of G color plane of all pixels are interpolated. Then, the values of G color plane are used as the gray values to obtain the gray cigarette image. Finally, According to the R and B color plane values, internal areas of cigarettes are smoothed and edges of cigarettes are sharped. As the aims are different, the ordinary methods that measure the ordinary interpolation methods are not effective to measure the interpolation method for cigarette counting. The gradient of all pixels are used to measure whether the interpolation algorithm render the cigarettes edges. Then, the concept of inner area smoothness and edge sharpness are brought forward to measure the interpolation method for cigarette counting.To effectively distinguish background and cigarettes in a cigarette image, there is a need to select a threshold to convert the interpolated image to the binary image. First, according to the geometric characteristics of cigarette container, the container areas in the image are erased. Then, as the gray values of pixels in the cigarettes are very different and the intervals between cigarettes are small, the whole image is divided into isometrical foursquare areas and the concept of area sparsity ratio is brought forward. The thresholds are selected according to the maximum entropy method in the areas with small sparsity ratio, while the thresholds are selected according to the average gray values in the areas with large sparsity ratio.In order to certifying the validity of the image acquisition method and binarization method, the above-mentioned methods are implemented in a PC. The experimental result proved that the interpolation method can clearly differentiate the background and the cigarettes in the image with the precondition that does not destroy the cigarettes geometric characterstics. At the same time, the thresholding method can effective segment the background and the cigarettes.