Static Body Feature – based Approach
The most popular approach to classify gender is through facial image. It is not obtrusive and suitable for automatic recognition application. (Bash et al., 2012) in their work suggested a gender recognition approach that involves using facial image, feature selection from each face was done by continuous wavelet transform and classification as male or female was done using SVM with linear Kernel. The major challenge with (Bash et al., 2012) approach is that the method they used for their feature extraction cannot resist the impact of complex backgrounds (Lin et al.
, 2016). The local feature extraction methodology extract features from specific facial points like the mouth, nose and eyes (Bing et al., 2012), while the global feature extraction method extract features from the whole face instead of extracting features from facial points (Caifeng, 2013).
Dynamic Body Feature based Approach
Gender classification based on static body features can perform identification. Since peoples appearances, styles, and locations changes often, then behavioural features like body movement and activity can be used for gender recognition.
Gender recognition approached with dynamic body features is more accurate than static body features because it incorporates features such as carriage and body language (Lin et al., 2016).
Though the method used to capture dynamic body feature is similar to that of static body feature using a camera, it however needs more continuous frames to capture the dynamic body features. This makes this gender classification method to require a higher computational complexity because behaviour features need image sequencing for recording movements.
Apparel Feature – based Approach
Apparel features is another source of features to perform gender classification, and it easier to obtain and discriminate even with low-quality images. Males and females have distinctive preferences concerning dressing. Hairstyle and clothing can be integrated in gender recognition.
2.4.2 Biological – based Gender Approach
Compared to the vision based approach, biological information does not change over a long period of time. Biometrics and bio signal information are examples of biological based approach for gender classification (Lin et al., 2016). Gender classification through biological information is more suitable for long term identification when compared to other non-invasive techniques like feature-based recognition because the biological based approach is significantly unaffected by aging. A gender classification approach through biometric information has a lower anti-tamper and distinctive trait than the classification approach through bio-signals (Lin et al., 2016).
Biometric Information – based Gender Classification
Biometric data are considered a better substitute that can be used in gender recognition because they are less affected by factors like mood and clothing. Biometric information used for gender classification are iris, fingerprint, voice, ear etc. Use of ear, iris and fingerprints for gender classification are based on image processing techniques. The data processing speed made them appropriate for a large scale identification system (Lin et al., 2016).
Voice and emotional speech are based on measurements of physical and behavioural attributes, the algorithm used for feature extraction uses global information and it requires a lot of computational resources. Because the algorithm is sensitive to background noise they are not suitable for a largescale classification system (Lin et al., 2016).