Tire Imprint Identification and Classification using VGG19

In the fields of forensic research, the automobile industry, and consumer safety, classifying tire imprints is a crucial task. It involves identifying the brand or producer of a tire based on the imprints left on various surfaces. Tire impressions can be valuable evidence in criminal investigations, counterfeit identification, product quality control, and accident reconstruction. Here, we propose a deep learning approach using the VGG19 architecture for tire impression identification and classification. To begin, we collected tire impressions from various two- and four-wheeler MRF tire brands. Our dataset consists of 2693 photos divided into 22 classes, with 15 classes representing 2-wheelers and 7 classes representing 4-wheelers. We processed the images through denoising, normalization, and enhancement steps. We then employed a mask-based segmentation method to extract texture-based information from the tire impressions. This, combined with morphological data, resulted in a comprehensive set of features used in pattern recognition. These features include statistical measurements such as mean, entropy, correlation, contrast, and others. To accurately determine the tire brand, we utilized the Structural Similarity Index Matrix (SSIM) with a minimum similarity index value of 0.7. By comparing the SSIM values, we were able to match the tire impressions to the correct brand. We divided the dataset into 80% training data and 20% test data, which provided the best performance for our proposed technique. The accuracy achieved using this approach was 95.3%.