Gender dictionary learning for gender classification

- A sparse dictionary learning approach is proposed for purpose of gender classification.
- Two separate dictionaries are proposed for male and female genders alongside with feature dictionary.
- Two dictionary learning methods are proposed to learn concurrent three dictionaries.
- A probability decision making approach is proposed to infer gender from their male and female probabilities.
- It improves the gender classification rate on the FERET, LFW and Groups databases.

Human gender is one of the important demographic distinctiveness for facial image description. In this paper, a novel method is proposed for gender classification from real-world images under wide ranges of pose, expression and so on. To this end, an automatic feature extraction method is proposed by two types of features. Then, two separate dictionaries for male and female genders are defined for representing the gender in facial images. Also, two dictionary learning methods are proposed to learn the defined dictionaries in training process. Then, the Sparse Representation Classification (SRC) is adopted for classification in the testing process. Finally, a probability decision making approach is proposed to classify the gender from estimated values by SRC and proposed gender formulation. Convincing results are obtained for gender classification on three publicity databases including the FERET, LFW and Groups databases compared to several state-of-the-arts.