Author(s)

Shasha Wu and Xueyun Chen

Corresponding Author

Shasha Wu

ABSTRACT

The low resolution and poor recognition of face image in the monitoring environment will lead to the decrease of face recognition accuracy. At present, most super-resolution algorithms suffer from serious image detail losses, due to the low-resolution of input images. In this paper, we propose a super-resolution reconstruction algorithm based on multi-scale feature fusion to alleviate this problem. A feature fusion mapping structure is used to extract features from multi-scale visual fields, and a skip-connection network structure is designed to construct high-resolution face images from them. The experimental results show that the proposed algorithm achieves better super-resolution results: clearer textures, sharper edges, enhanced visual effects, and higher evaluation indexes on FERET face database than the existing mainstream algorithms.

KEYWORDS

Face super-resolution reconstruction; multi-scale; feature fusion; convolutional neural network; image processing