Abstract
Automatically scene understanding using machine learning algorithms has beenwidely applied to different industries to reduce the cost of manual labor.Nowadays, insurance companies launch express vehicle insurance claim andsettlement by allowing customers uploading pictures taken by mobile devices.This kind of insurance claim is treated as small claim and can be processedeither manually or automatically in a quick fashion. However, due to theincreasing amount of claims every day, system or people are likely to be fooledby repeated claims for identical case leading to big lost to insurancecompanies.Thus, an anti-fraud checking before processing the claim isnecessary. We create the first data set of car damage images collected frominternet and local parking lots. In addition, we proposed an approach togenerate robust deep features by locating the damages accurately andefficiently in the images. The state-of-the-art real-time object detector YOLO\cite{redmon2016you}is modified to train and discover damage region as animportant part of the pipeline. Both local and global deep features areextracted using VGG model\cite{Simonyan14c}, which are fused later for morerobust system performance. Experiments show our approach is effective inpreventing fraud claims as well as meet the requirement to speed up theinsurance claim prepossessing.