Face data pair-oriented complete face reticulate pattern restoration method and system and medium

Abstract

The invention discloses a face data pair-oriented complete face reticulate pattern restoration method and system and a medium. The method comprises the following steps of combindinga face reticulate pattern image and a clean face image corresponding to the face reticulate pattern image into a picture in the horizontal dimension; taking a training set composed of the data pairs as input of a GAN network, and training to obtain a restoration model; and inputting a to-be-restored face reticulate pattern image into the restoration model to obtain a restored clean face image. The method is an end-to-end training mode, the step of reticulate pattern recognition is omitted, and the bottleneck of reticulate pattern segmentation-based restoration is broken through. The image with the reticulate pattern is used as noise data, the face image without the reticulate pattern can be directly generated through end-to-end adversarial training, the integrity of the restored face reticulate pattern imageis improved while the face reticulate pattern restoration process is simplified, and obvious restoration traces cannot appear.

Description

technical field [0001] The invention relates to the field of face image processing, in particular to a method, system and medium for face texture restoration for complete face data. Background technique [0002] The existing image style conversion algorithms all need to be initialized, and the parameters of the CNN need to be fixed before backpropagating to update the image, which has poor performance. [0003] The Pix2pix network has a good performance when doing image restoration. Use the clean face and face texture data as the input data pair, where the face texture is equivalent to the noise data, the clean face is equivalent to the target image, and the image generated by the generation network is the output image. The Pix2pix generation network uses the U-net structure with Encoder-Decoder, which allows each deconvolution layer to have the features extracted by the downsampling convolution layer, so that it can carry more information. Its discriminative network also ...

AI Technical Summary

Problems solved by technology

If this multi-task facial texture repair method does not segment the precise texture in the first step, it will affect the subsequent texture facial restoration task, and there will be obvious repair traces.

In a separate segmentation or repair task, it is necessary to convert the size of the image first, and finally convert it to the target resolution, which will lose the original information of the image and cause the repaired image to be distorted

There is still a big gap between the details of the face image generated in the traditional mesh face image restoration method and the original image

Images