Seal denoising identification method

The invention provides a seal de-noising identification method, which comprises the following steps: obtaining a semantic segmentation network model through ResNet-based Unet network training, carrying out de-noising processing on a seal image with a contract bill background by using the semantic segmentation network model, carrying out feature extraction on a reserved seal image through a convolutional neural network, and establishing a reserved image feature database; and identifying by comparing the features of the de-noised seal image. According to the seal de-noising identification method, the influence of factors such as non-uniform seal stamping force of a client and the color depth of seal inkpad on a seal identification result is reduced, and the influence of factors such as characters, shading, signatures and tables together with the seal on the seal identification result is reduced; and the influence of factors such as lens pollution, illumination and shooting angle on the seal identification resul