Open AccessSmall Face Detection Based On Feature Fusion
Author(s) -
Zili Hao,
Hangqi Shi
Publication year - 2021
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1871/1/012086
Subject(s) - computer science , face detection , artificial intelligence , face (sociological concept) , feature (linguistics) , convolution (computer science) , computer vision , detector , pattern recognition (psychology) , pixel , object class detection , upsampling , facial recognition system , image (mathematics) , artificial neural network , telecommunications , social science , sociology , linguistics , philosophy
While great progress has been made in face detection, one of the remaining challenges is to detect the faces of small targets in images. The small target face we define, mainly includes the absolute size of the face is not greater than 32×32 pixels and the relative proportion of the face size is not greater than one tenth of the image size. To meet this challenge, we propose an effective face detector, called SmallFaceBoxes, which has a better performance in both speed and accuracy. Specifically, our method has a lightweight but powerful network structure consisting of small target face detection layers and multi-scale convolution layers. The design of small target face detection layers enables detector to increase the representation ability of small and medium-sized faces in the image and improve the detection accuracy of small faces. The multi-scale convolution layers aim at enriching the receptive fields and discretizing anchors over different layers to handle faces of various scales. In addition, according to the size of receptive field, we propose a strategy of setting Anchor in multiple feature layers, thus improving the recall rate of small face. As a consequence, our algorithm performs well on the wideface dataset.