Mask-shielded face recovery method based on an adaptive context attention mechanism

Abstract

The invention discloses a mask-shielded face recovery method based on an adaptive context attention mechanism. The method comprises the following steps: downloading a trained mask segmentation full convolutional neural network FCN and an untrained U-shaped network; respectively constructing an image fine recovery network and a global and mask shielding area discrimination network; collecting a training set and a test set of paired mask-shielded and non-shielded face images; inputting the mask shielding image into FCN to obtain a mask mask; sequentially training a U-shaped network, an image fine recovery network and a global and mask shielding area discrimination network; training the image fine recovery network again by using the output of the global and mask shielding area discrimination network; and inputting the mask shielding image in the test set into the trained U-shaped network, and inputting the result and the mask mask into the re-trained image fine recovery network to obtain a fine recovery image. The recovered image is consistent in color and texture, the image is clear, and the method can be used for face detection and face recognition.

Description

technical field [0001] The invention belongs to the technical field of computer vision, and in particular relates to an occluded face restoration method, which can be used for face detection and face recognition. Background technique [0002] Face restoration is an important task in computer vision. Its purpose is to fill in the missing areas of occluded faces. It has a wide range of applications in occluded face detection and occluded face recognition. In recent years, most deep learning based face restoration methods have achieved remarkable results. These methods usually use state-of-the-art network architectures such as U-Net, or design new loss functions such as reconstruction loss to restore occluded facial images. However, due to the problems of variable face poses and various occlusions, the quality of restored images obtained by existing face restoration methods is still not satisfactory. [0003] With the outbreak of respiratory infectious diseases, more and more...

AI Technical Summary

Problems solved by technology

There are two shortcomings of this method: one is that the context-related information between the occluded area and the unoccluded area of ​​the image is not used, which will cause the problem of color and texture inconsistency between the restored image of the occluded area and the unoccluded area of ​​the image

The second is that it is only effective for small-area occlusions, and the recovery effect for large-area occlusions such as masks is poor, and the image in the restored area is blurred and has artifacts

Images