Loading...
| Friend's email | |
| Your name | |
| Your email | |
| enter code | |
This page was sent successfuly
- Type of Document: M.Sc. Thesis
- Language: Farsi
- Document No: 51372 (05)
- University: Sharif University of Technology
- Department: Electrical Engineering
- Advisor(s): Ghaemmaghami, Shahrokh
- Abstract:
- Nowadays, using of digital cameras is streaming across the world dramatically. Application of these devices is very diverse. One of the most interesting application of those is face verification. For example, imagine your smartphone has an application which verifies faces in front of its front camera, if that face be your face (with variation from original) then application automatically unlocks your phone. Face verification systems are also deployed in airports to verify passport photos and in smart homes. One of the most regular problems in face verification is occlusion. When your face is occluded with natural or random changes we can say your face is occluded. All of the recent papers about the face verification use Deep Learning (DL) as a main part for verifying faces. Based on this fact, we first review the some recent modern papers introducing DeepFace,FaceNet,VGGFace, OpenFace,ResNet and DeepID networks for face verification with DL approach and then we test one of them with inputs occluded under different values. After this test, we detect the problem that exists in this network and try to introduce a cure for that problem. Our proposed approach is based on augmented images of training data around the center of faces. Then, we propose two models for verifier network to use in this thesis. First model is based on Convolutional Neural Network which we train it from scratch. The main problem in this aproach is inadequate training set which limited computational performance is cause of this problem. Second, is a method based on the Transfer Learning (TF) which uses extracted features from a pretrained network. We use this approach and try to train our face verification deep network with these extracted features. Simulation results show that the proposed method outperforms the original pretrained network under occluded inputs. Under 50 percent of occlusion our proposed network verifies faces with accuracy of 74:1 percent while it is 68:4 in the FaceNet network. Several proposals are mentioned in the final part of the thesis for future research in this area
- Keywords:
- Deep Learning ; Occlusion ; Feature Extraction ; Identity Verification ; Face Verification
Digital Object List-
محتواي کتاب - view
Bookmark