Gender recognition using four statistical feature techniques: a comparative study of performance

• Published in: Evolutionary intelligence Vol. 12; no. 4; pp. 633 - 646
• Main Authors: Al-wajih, Ebrahim, Ghouti, Lahouari
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.12.2019; Springer Nature B.V
• Subjects: Applications of Mathematics; Artificial Intelligence; Bioinformatics; Chronology; Comparative studies; Control; Discrete cosine transform; Engineering; Feature extraction; Feature recognition; Gender; Histograms; Mathematical and Computational Engineering; Mechatronics; Organs; Performance enhancement; Principal components analysis; Research Paper; Robotics; Statistical Physics and Dynamical Systems; Gender recognition; Pyramid histogram of oriented gradients; Biometric system; Gabor filters; Statistical features
• ISSN: 1864-5909; 1864-5917
• DOI: 10.1007/s12065-019-00264-z

Nowadays, many applications use biometric systems as a security purpose. These systems use fingerprints, iris, retina, hand geometry, etc. that have unique patterns from person to another. The human face is one of the most important organs that has many physiological characteristics such as the subject gender, race, age, and mood. Determining the gender of the face can reduce the processing time of large-scale face-based systems and may improve the performance. Many studies were proposed for gender recognition, but several were evaluated using the accuracy as a performance metric which is improper for unbalanced data. Further, they used a grayscale color; and extracted features either from the whole image or equally divided blocks, as a grid. In this paper, novel methods are proposed based on statistical features that have the ability to represent the face landmarks. These features are GIST, pyramid histogram of oriented gradients, GIST based on discrete cosine transform and principal component analysis that are extracted using face local regions. The performances are evaluated using area-under-the-curve that is computed from the receiver operating characteristic or ROC curve. At the end, the acquired performance has been compared by two state-of-the-art techniques that shows that the proposed approaches enhance the performance between 1 and 3%, but the number of features is increased.