A large number of approaches for gender classification use Suport Vector Machine. In 24 ahigh accuracy gender classification approach using SVM on low resolution thumbnail imageswas reported. Lian and Lu in 25 presented multi-view gender classification by trainingSVM on histograms of Local Binary Patterns given head pose as global descriptor.
Lesscomputationally expensive method was presented in 26, where Adaboost learning methodwith set of Look Up Table (LUT) weak classifiers trained on rectangular features was used.Xu and Huang 27 used Adaboost framework as well, with Second Order DiscriminantAnalysis (SODA) classifiers were used to construct strong classifier. Prior to development ofage estimation methods, a number of studies were focused on simulation of age effects.
In28 simulation of aging facial feature was performed by superimposing some characteristicshape and colour changes. In 29 statistical method that aims to model historical, familialand average aging tendencies of the peer group was reported. Although those methods areinverse of age estimation process, age simulation methods can be used in order to findimportant dependences between age and face that would later facilitate system for ageestimation. One among first works addressing issue of age estimation was presented by Kwonand Lobo in 30, where facial images were classified into three groups according to thecombination of facial features of different natures, in particular features based on distributionof facial feature points and features that are able to capture texture of the skin (wrinkles).Method proposed in 31 model age features which are represented as sequence of faceimages across the time by defining suitable subspace. Lanitis et al. 32 used ActiveAppearance Models for face encoding, and using an aging modelling function and largenumber of training images they were able to capture dependence between encodedinformation about face and actual age of the subject.