Vehicle Damage Classification and Fraudulent Image Detection Including Moiré Effect Using Deep Learning

Abstract

Image-based vehicle insurance processing and loan management has large scope for automation in automotive industry. In this paper we consider the problem of car damage classification, where categories include medium damage, huge damage and no damage. Based on deep learning techniques, MobileNet model is proposed with transfer learning for classification. Moreover, moving towards automation also comes with diverse hurdles; users can upload fake images like screenshots or taking pictures from computer screens, etc. To tackle this problem a hybrid approach is proposed to provide only authentic images to algorithm for damage classification as input. In this regard, moiré effect detection and metadata analysis is performed to detect fraudulent images. For damage classification 95% and for moiré effect detection 99% accuracy is achieved.