Classification modeling method for hyperspectral stamp-pad ink data based on one-dimensional convolutional neural network

Assessment of high-quality counterfeit stamp impressions generated by inkjet printers via texture analysis and likelihood ratio

High-quality counterfeit stamp impressions made by inkjet printers remain challenging in questioned document examination and forensic analyses. A dataset comprised of various printed stamp impressions, using ten options of conditions and materials, and hand stamped impressions was generated. In this paper, we report printed impressions in pure color and high-quality printing mode are very similar to hand stamped impressions in terms of their microscopic characteristics. These similarities may lead to incorrect conclusions via traditional identification methods. Here, we proposed a method for identifying counterfeit stamp impressions via texture features and image quality parameters extracted from impressions. First, the statistical analysis methods were used to verify a significant difference between the printed and hand stamped impressions. Principal component analysis (PCA) was used to show the variation between the impressions, and the differences between printed and hand stamped impressions were obvious in the three-dimensional plot. After filtering the background of the stamp impressions, image processing analysis was introduced to extract features of gray level co-occurrence matrix (GLCM), segmentation-based fractal texture analysis (SFTA), local binary pattern (LBP), and image quality metrics (IQM), which were used to characterize the stamp impressions. Finally, specific cases were simulated by random selection, based on the dataset of stamp impressions, and an evaluation system for stamp evidence was established to calculate the likelihood ratios (LRs) under two alternative hypotheses. The likelihood ratio interprets calibrated evaluations on the strength of stamp impressions as evidence. We can also balance these LRs against the rates of misleading evidence with a reasonable performance (equal error rate = 0.048). This paper provides a system to differentiate high-quality printed and hand stamped impressions with reasonable performance.