The work aims to filter the noise from the engine surface defect image so that the engine surface defect information can be better presented. Firstly, wavelet transform was used to decompose the noise image of engine surface defect to obtain different wavelet coefficients. Then, SVM was used to classify the wavelet decomposition coefficients in order to separate the noise signal from the non-noise signal. Finally, interpolation was used to optimize the hard threshold function to overcome the ringing effect caused by the discontinuity of the function, so that the denoised image could keep more details. The denoising effect of the proposed method was compared with that of the median filter and the bilateral filter through experimental simulation. The image denoised by the proposed method had higher PSNR and SSIM values than that denoised by median filtering and bilateral filtering methods. When the noise of tested image was 25%, the PSNR value and SSIM value of the image denoised by the proposed method increased by 20.66% and 11.89% respectively, compared with that denoised by the median filtering method and increased by 10.30% and 5.48% respectively, compared with that denoised by bilateral filter. Therefore, the proposed method has better denoising effect than the median filter and bilateral filter. It can remove the noise of the engine surface defect image and retain the details of the image better.