Face recognition under diverse condition using deep learning

Eng, Xing Xiong (2017) Face recognition under diverse condition using deep learning. Project Report. Universiti Teknikal Malaysia Melaka, Melaka, Malaysia. (Submitted)

Text (Full Text) Face recognition under diverse condition using deep learning.pdf - Submitted Version Download (1MB)

Abstract

The project is focussing on analysing face recognition under diverse condition using deep learning. Face recognition is getting common in recent years, it can be apply on various application. Different situation means that there are times where face is not directly shown when recognizing a face, for example the subject is showing a side of his face or wearing a mask. The problem is hard to obtain good accuracy on this situation. Moreover, in order to train a model to recognize a face, the facial data are often need to select carefully such as an open face. Hence the purpose of this project is to apply deep learning which can improve different situation face recognition performance and to test the face recognition using deep learning technique. In this project a Convolutional Neural Network (CNN) is used to build the model, where CNN is known to have high accuracy in image recognition. The methodology used in this project is Evolutionary Prototyping Model which is suitable for this project because it has steps that repeatedly reviewing and refining in order to produce a desire product. Besides, Python programming language is used together with Google deep learning framework: Tensorflow to code the CNN. Because the training of a CNN is consuming quite a bit of computational power and time, therefore a GPU is used to speed up the process of the training process. The system consist of a simple GUI which allow user to select an image file (Portable Graymap Format, .pgm) as an input to the model, and it will display the similarity of the input image and trained image. The result of the system is good on some specially prepared data for testing, it is found that noise level can effect the result significantly. In conclusion, this project is a simply application of deep learning technique on face recognition task, and it gives a rough idea on how it perform in such task.

Actions (login required)

View Item