Fully convolutional Deep Stacked Denoising Sparse Auto encoder network for partial face reconstruction. (October 2022)

• Record Type: Journal Article
• Authors: Dinesh, P.S.; Manikandan, M.
• Is Part Of: Pattern recognition. Volume 130(2022)
• Publication Date: 2022-10
• Subjects: Partial face recognition -- Deep learning algorithm -- Fully convolutional network -- Autoencoder; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2022.108783
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: In this partial face detection (PFD) is used to overcome the challenges involved in face detection and reconstruction. A novel PFD algorithm called Self- motivated feature mapping (SMFM) combining a FCN and DS-DSA algorithm. The proposed system focuses the feature maps from the FCN and used by the DS-DSA to perform partial face reconstruction. The spatial maps are generated by extracting the features from FCN and supplied as the input for partial reconstruction. By using principle component analysis and linear regression method and re-identification to the DS-DSA network. Abstract: Face recognition is one of the most successful applications of image analysis. Since 1960s, automatic face recognition research has been carried out, but the problem is still unresolved. Therefore, in this manuscript, a novel Partial face reconstruction (PFR) algorithm called Self- motivated feature mapping (SMFM) combining a Fully Convolutional Network (FCN) and Deep Stacked Denoising Sparse Autoencoders (DS-DSA) algorithm is proposed to overcome the challenges. The proposed approach focuses on the generation of feature maps from the Fully Convolutional Network and it is used Deep Stacked Denoising Sparse Autoencoders to perform the partial face reconstruction. The spatial maps are generated by extracting the features from Fully Convolutional Network and it is supplied as the input for partial reconstruction and re-identification to the Deep Stacked Denoising Sparse Autoencoders network. The main aim of the proposed work is "to enhance the accuracy during facial reconstruction". The proposed approach is implemented in MATLAB platform. The performance of the proposed approach attains 23.45% and 20.41% accuracy, 25.93`% and 19.43% sensitivity, 22.21% and 24.41% precision and20.21% and 23.41% Specificity greater than the existing approaches, like Partial Face Reconstruction using generative adversarial networks (GANs), Partial Face Reconstruction using Deep Recurrent neural network (DRNN).