Image recognition based on deep learning of convolutional neural networks

• Main Author: Xie, Cong
• Other Authors: Jiang Xudong
• Format: Final Year Project
• Language: English
• Published: 2019
• Subjects: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Image processing and computer vision
• Online Access: http://hdl.handle.net/10356/77415
• Institution: Nanyang Technological University

Face recognition is a new process to identify people in some resource which are from images or videos. The face is becoming an important factor to recognizing people in modern and social life. Algorithms of face recognition system is to extract facial features, then make comparison on those with a database to find the whether they are matched. [1] Convolutional Neural Network (CNN) is an important method architecture of Deep Learning. CNN has achieved impressive results in the area of image classification. PCA algorithm is another facial recognition method which mainly focuses on pretreatment of face image location, input of face database, extracting face feature and face recognition four parts. There is a statistical method called PCA, which is used to make the number of variables in face recognition reduced. [2] Every image of the training group is expressed to be a linear combination of weighted feature vectors. Those can be called to be feature faces. These eigenvectors can be obtained from the covariance matrix of the training image group. After electing the most relevant group of feature faces, can find the weights. Recognition progress is that projecting the test image onto the subspace spanned by the feature face, and through calculating the minimum Euclidean distance to carry out classification. To test face recognition system performance through different experiments. This project focuses on compare recognition accuracy performance of Convolutional Neural Network (CNN) with Principal Component Analysis (PCA). The project two algorithms progress and experiment based on ORL face database which contains 400 images of size 112 x 92. And overall images composed of 40 persons and 10 images per each person. [2]