Abstract:
Humans can effortlessly determine the gender of other person. This has stimulated interest to enable computer machine of accurate guessing the human faces as male or female. Major problems in face classification are due to the large variance in appearance in a digital image when it is captured / exposed to different lightning and unfamiliar pose. Gender classification has an extensive usage in numerous problems and domains. Automated gender classification is an area of great significance and has huge impact and potential for future research. Its use is significant in several industrial applications such as monitoring, security, surveillance, biometrics, commercial profiling and human computer interaction. Gender has been identified using different traits like gait, iris and hand shape but a major and significant work has been carried out based on face. Main emphasis of this research work is critical assessment of different methods used in Gender Classification and highlighting favorable and unfavorable factors of these existing techniques. In next sections methodologies have been presented for efficient gender classification in still images and animated videos and over smart phones. Schemes have been presented for these diverse medium of digital image processing. We have conducted experiments to identify gender for the comparisons purpose for both areas of focus i.e. consumer face images captured run-time and fictional characters in animated movies. Flow of work, implementation of proposed classification methodology and learning algorithm is part of this thesis.
Main modules of Gender Classification task are image acquisition, face detection, image normalization, feature extraction and classification. Every task has been thoroughly
iii
investigated with state of art methods and then final modeling is proposed, implemented and tested.
