The work aims to propose a method based on visual word bag model to extract the features of plastic marks in cigarette packets to quickly and accurately identify the authenticity of multi-brand cigarettes. Firstly, a self-designed multi-light source glue mark acquisition device was used to obtain the glue mark image inside the cigarette packet, and the glue mark image sample was obtained after removing part of the background of the original image by image processing technology. Then, scale invariant Feature conversion (SIFT) features were extracted from the glue mark image samples, and K-Means algorithm was used to cluster the features to generate a visual dictionary. Then, according to the visual word histogram feature set of the visual dictionary, the glue mark images were trained and classified, so as to identify the authenticity of cigarette. In this paper, 10 samples of authentic cigarette packets and counterfeit cigarette packets of 64 cigarette brands were taken as the objects. The classification test of 360 cigarette packet images showed that the authenticity recognition rate was 97.22%, and the average identification time of each sample was less than 0.05 s. The above method is simple to collect glue marks, has high authenticity identification efficiency and accuracy, and is suitable for a variety of cigarette brands. It provides technical support for improving the efficiency, accuracy and universality of authenticity identification.