Method of tire pattern image retrieval based on wavelet transform and Siamese network
Yueya Qiang, Chiyun Sheng, Dong Yin · Proceedings of the 2020 International Conference on Aviation Safety and Information Technology · 2020
Image retrieval of tire tread pattern plays a vital role in traffic accidents and criminal investigation. We propose a tire pattern retrieval method based on wavelet transform and Siamese network, which combine traditional techniques with deep learning. Firstly, we use the wavelet transform to extract low-level features of tire patterns. Secondly, we design a lightweight network called TireNet to extract high-level features of tire patterns. The TireNet based on MobileNet V1 replace the single convolutional neural network of the original Siamese network. Finally, the low-level and high-level features are fused to generate a one-dimensional vector, and calculated the similarity by Euclidean distance. To verify the proposed method in this article, we construct a tire pattern dataset named TireDataset. The experimental results show that the Siamese network is feasible in the task of learning the similarity of tire patterns, and TireNet outperforms existing methods in terms of speed and accuracy trade-off.