| dc.description.abstract | Gender is a very important personal attribute inherent in human beings. Hence, an automatic and effective gender recognition system is desirable in various applications, such as intelligent surveillance system, human-computer interaction, and customer behavior analysis. Take customer behavior analysis for example, the applying of gender recognition technology in digital signage can attract potential customers while demonstrating custom advertisements.
Since human face conveys a clear sexual dimorphism, the using of facial features seems to be an intuitive way for gender recognition. In consequence, abundant gender recognition methods were proposed based on certain facial components. However, people may wear sunglasses or scarves in real life which will result in occluded facial image. Under these circumstances, the performance of gender recognition will thus be degraded due to the occlusion problem. In this thesis, we propose an adaptive gender recognition method by dynamically selecting facial components to cope with the aforementioned problem. In the occlusion classification procedure, the occluded facial components are firstly detected and only the non-occluded facial components are selected in the recognition role to overcome the failure problem suffering the recognition performance.
Three different experiments were conducted to verify the validity of our proposed method. They were categorized in terms of single classifier of components, multiple classifiers for adaptive gender recognition, and system robustness test by using partial occlusion image. Experimental results demonstrate that the proposed method exhibit excel accuracy under various conditions. | en_US |