Fractal Anti-Counterfeit Label Comparison Using Combined Image Features and Clustering

The comparison of traditional fractal anti-counterfeit labels is based mainly on manual inspection. However, such labels have rich details and complex structures, making the entire identification process labor-intensive. Thus, manual inspections are highly susceptible to low identification accuracy, which produces unreliable results. To best address these disadvantages, an automatic comparison method for fractal anti-counterfeit labels is proposed. The method can effectively extract the color features, texture features, and shape features of anti-counterfeit labels, and perform cluster analysis and comparison. First, a color volume histogram is used to extract the color and pixel space information features from the fractal anti-counterfeit labels. To compensate for the deficiency of using a single feature, texture and shape features were also extracted based on the median robust extended local binary patterns (MRELBP) and Hu moments. Next, based on feature extraction, k-means clustering is performed to ensure that as many of the same types of labels as possible can be divided into the same clusters and that the comparison can only be performed on one or a few clusters. The experimental results show that the speed and quality of fractal anti-counterfeit label comparison effectively improved after the clustering analysis. Furthermore, based on visual identification and unreliable comparison results, the proposed method is expected to help consumers quickly identify low-quality, counterfeit products.