An adaptive face recognition under constrained environment for smartphone database

Face recognition is probably one of the most prominent areas of imaging research and has a wide range of real-world applications. Although face recognition has recently achieved advances in identifying people, limitations and challenges remain in face recognition applications in which no restriction is imposed on the conditions of acquired facial videos. This thesis is concerned with face recognition under uncontrolled environments in which the images used for training and testing are collected from the real world using a smartphone camera. For now, publicly available smartphone face databases remain lacking. In addition, existing databases do not address all the challenges of real-world scenarios. One of the crucial problems in the uncontrolled environment of smartphone data is illumination variation, which negatively affects the preservation of image features caused by binary conversion. In addition, using data from smartphone devices introduces a new challenge, namely, different optical zooms. This problem affects the accuracy of face recognition systems when the test and gallery images of the same person differ in terms of face-to-camera distance. Moreover, the performance of recently developed face detection methods is poor under uncontrolled environments, such as those with variations in illumination, complex background and overlapping between face and background colour. In fact, detecting the correct face boundary is insufficient to extract the correct features of the face region, particularly in the presence of occlusion, which affects the feature extraction operation and decreases the accuracy of face recognition. Finally, increasing the accuracy of a face recognition method under the complex environment of a smartphone face database remains a considerable challenge among researchers. The first objective of this study is to construct a smartphone face video database that closely reflects real-world videos. The next objective is to enhance the appearance of face features under various illumination conditions by converting an image into a binary image using a new columnar binary conversion method considering the robustness and strong discriminative power of binary features. In addition, this study aims to improve the performance of the face recognition method under the effect of different optical zooms by detecting the normalised facial feature region using the proposed facial feature region normalisation method. The next objective is to accurately detect the face region under a complex background and varying illumination conditions using the proposed geometric skin colour method. Furthermore, this thesis aims to solve the problem of unintentional occlusion by detecting the correct non-occluded area using the non-occluded facial area detection method. Finally, this study aims to obtain high-accuracy face recognition performance under the uncontrolled environment of a smartphone database based on the proposed adaptive face recognition method that combines two new face recognition algorithms. The proposed method works by adapting to the environment of the input image and uses multiple facial features to increase the reliability and efficiency of the recognition process. Experimental results showed that the recognition rate achieved 100% under different environment conditions.

Text FSKTM 2018 71 - IR.pdf Download (2MB)

Actions (login required)

View Item