An improved multi-scale face detection using convolutional neural network
- PDF / 1,436,843 Bytes
- 9 Pages / 595.276 x 790.866 pts Page_size
- 84 Downloads / 235 Views
ORIGINAL PAPER
An improved multi-scale face detection using convolutional neural network Hazar Mliki2 · Sahar Dammak1 · Emna Fendri3 Received: 7 August 2019 / Revised: 11 December 2019 / Accepted: 19 March 2020 © Springer-Verlag London Ltd., part of Springer Nature 2020
Abstract In this paper, we introduce a deep learning (CNN) based method for face detection in an uncontrolled environment. The proposed method consists in developing a CNN architecture dedicated to the face detection tasks by combining both of global and local features at multiple scales. Our architecture is composed of two main networks: A region proposal network that generates a list of regions of interest (ROIs) and a second corresponds to a network that use these ROIs for classification into face/non-face. Both of them share the full-image convolution features of a pre-trained ResNet-50 model. Experimental study was conducted on the famous WIDER Face and FDDB databases. The obtained results proved the efficiency as well as the feasibility of the proposed method to deal with multi-scale face detection problems. Keywords Face detection · CNN · Transfer learning · NMS
1 Introduction In recent decades, Several research [4,5,16,21] works have been particularly oriented towards the field of detection and analysis on human subjects using facial features. This refers basically to their usefulness in terms of multiple applications, such as biometrics for access control, video surveillance systems and other security applications. However, to perform face analysis tasks, face detection needs to be achieved. Several studies [1,5,10,14,22,24] have explored the issue of face detection for decades. Yet, despite the significant progress recorded in the field, robust face detection in an uncontrolled environment remains a relatively obscure area owing to the appearance of large variation of faces affected by occlusion, pose variations, low resolutions, scale variations, illumination variations, etc. In this research paper, we set forward a new method for face detection based on convolutional neural networks (CNN). In fact, the proposed method introduces a new CNN architecture that extends the Faster R-CNN [17] architecture by combining both global and local features at multiple
B
Sahar Dammak [email protected]
1
University of Sfax, MIRACL-FSEG, Sfax, Tunisia
2
University of Sfax, MIRACL-ENET’COM, Sfax, Tunisia
3
University of Sfax, MIRACL-FS, Sfax, Tunisia
scales. The proposed method consists of three steps: (1) feature extraction using a pre-trained model, (2) region of interest (ROI) generation and (3) region of interest (ROI) classification into face/non-face. The main contributions of this work can be sum up as follows: – In the feature extraction step, we used the ResNet50 pretrained model [7]. Its merit resides in the fact that it integrates residual blocks that extract local and global features. – In the region of interest (ROI) classification step, we applied a combination of multi-scale feature maps to enhance the feature vectors e
Data Loading...
