Detection of recaptured vehicle images in automotive repair scenarios using convolutional neural networks

In the automotive repair industry, the authenticity of photographic evidence is critical for insurance claims, remote damage assessment, and repair records. Recaptured photos (photographed from screens rather than original scenes) may obscure details, introduce artifacts, and even be used for fraud. In this paper, we propose a novel recaptured vehicle image detection method based on a specially designed multi-input convolutional neural network (CNN) architecture. We construct a dedicated vehicle image database containing original photos taken directly with mobile devices, as well as images recaptured from original photos displayed on various screens (using different smartphones), fully simulating a typical automotive repair environment. Experimental results on this dataset show that our method can distinguish between original and recaptured photos with an accuracy of 96.8%, significantly outperforming traditional analysis methods in automotive repair scenarios.