Face illusion structure method and system based on double-path deep fusion

Abstract

The invention discloses a face illusion structure method and system based on double-path deep fusion, and the method comprises the steps of extracting human face features through cyclic convolution and residual learning in a face global feature extraction process, and carrying out the global contour modeling of a face; in a block-based face local structure information expression process, samplingan input face to obtain a large number of local face blocks for independently modeling a mapping relationship between low-resolution and high-resolution local blocks; in addition, rearranging and mapping the feature expressions of the local blocks back to the global face space; in the face global and local feature fusion illusion structure, fusing the global contour features and the local featuresof the face, so that the high-quality face illusion structure is realized. According to the invention, the global contour features of the human face are effectively extracted, the high-dimensional and low-dimensional mapping relation of the local blocks is independently learned, effective fusion is achieved, the contour of the illusion human face image is clearer, and texture details are richer.

Description

technical field [0001] The invention belongs to the technical field of digital images, and relates to a low-resolution human face illusion construction method and system, in particular to a human face illusion construction method and system based on dual-path deep fusion. Background technique [0002] Low-quality face images obtained by video surveillance have the characteristics of less information, blurred content, and low resolution, which greatly affects the understanding of image or video content, thereby interfering with the accuracy of downstream high-level computer vision tasks. Therefore, low-resolution face illusion is a basic process to improve the visual experience of images, and it is also an important preprocessing step for many computer vision tasks (such as face detection, recognition, and pedestrian tracking, etc.). [0003] Since deep learning, especially convolutional neural network (CNN) has achieved good results in the field of image processing, in recen...

AI Technical Summary

Problems solved by technology

Therefore, for faces with extremely low resolution or compression distortion, existing face illusion algorithms usually have limited ability to represent local facial features, and cannot create high-quality face images.

Images