Automotive paint is one of the common physical evidence at the scene of traffic accidents. Rapid inspection of automotive paint can provide clues and evidence for case investigation. Optical coherence tomography (OCT), which has the advantages of strong tomography, has significant advantages in vehicle paint inspection, but noise interference will weaken the signal. In order to improve the recognition ability, this paper collected A-scan signals of 102 vehicle paint samples from 6 different brands, and three kinds of fast Fourier transform filters and Hilbert transform filters were used to denoise the original data of vehicle paint samples. Two classification models, support vector machine and Fisher discriminant analysis, were constructed to identify different brands of paint samples. The filter with the best noise reduction effect and the optimal classification model were selected according to the recognition accuracy and overall classification accuracy of different brands of paint samples. Results show that the overall classification accuracy of sample data processed by Hilbert transform filter and fast Fourier transform low-pass filter is significantly improved, and Hilbert transform filter is better than fast Fourier transform filter. Moreover, the overall classification accuracy of Fisher discriminant analysis model is higher than that of support vector machine model. The Fisher discriminant functions based on Hilbert transform filter realize the accurate recognition of six brand car paint samples, and the overall classification accuracy reaches 100%. The preprocessing method based on filter can effectively suppress the noise interference in the sample data, and greatly improve the efficiency and ability of optical coherence tomography in examining vehicle paint evidence.