An Hybrid Online Training Face Recognition System Using Pervasive Intelligence Assisted Semantic Information
In face recognition, the large sizes of training databases can place a heavy burden on computing resources and may produce unsatisfactory results due to significant amount of irrelevant features for target screening. We adopt the technology of wireless se
- PDF / 606,528 Bytes
- 7 Pages / 439.37 x 666.14 pts Page_size
- 66 Downloads / 191 Views
)
Department of Computer Science, University of Hull, Hull, England [email protected], {Y.Cheng,P.Jiang,Martin.Walker}@hull.ac.uk
Abstract. In face recognition, the large sizes of training databases can place a heavy burden on computing resources and may produce unsatisfactory results due to significant amount of irrelevant features for target screening. We adopt the technology of wireless sensor networks by storing semantic information in wire‐ less tags to assist grouping of candidates. The semantic information of nearby people such as gender and race is provided to the robot and help it narrows its search to a smaller subset of the database. Hence the face recognizer can be simplified by training the selected subset samples that makes online training possible. Furthermore, the feature space can be constantly adjusted benefiting from online training to distinguish faces with higher accuracy and the resolution of training samples can also be adjusted based on the camera and target distance. In order to further improve the correct rate, permutation post processing has been employed. The proposed hybrid approach has been validated in experiments with a promising low error rate. Compared to other face recognition systems, our system is better suited to work on a human-machine interactive robot which needs to detect targets under different illumination conditions and from different distances. Keywords: Face recognition · Semantic information · Pervasive intelligence
1
Introduction
The size of a human’s social network is estimated to be 150 [1]. It is suggested that the number of neocortical neurons in our brain limits the organism’s information-processing capacity, and this then limits the number of relationships that an individual can monitor simultaneously. So, even if a person claims he or she has thousands of friends on Face‐ book, only about 150 of those relationships are meaningful in their head [2]. Though sometimes we are bothered by bad memory, few of us like to record a person’s information in a contact book in detail. People are used to describing each other with more abstract words, such as gender, race, age, color of hair, etc. This semantic information is helpful and can easily guide us in picking a person out of a crowd. Just as people use a contact book, a robot has all information stored in its database. A face recognition system is a computer application that automatically verifies an © Springer International Publishing Switzerland 2016 L. Alboul et al. (Eds.): TAROS 2016, LNAI 9716, pp. 364–370, 2016. DOI: 10.1007/978-3-319-40379-3_37
An Hybrid Online Training Face Recognition System
365
identity from a digital image or a video frame. The system recognizes faces by comparing the selected facial features from a facial database [3]. Though large memory is not a big problem for a robot today, it still cannot recognize faces as well as we do. That is because the world model installed in the robot cannot be completely consistent with its observed real world. The results of face recognition are easil
Data Loading...
