A Robust Feature Selection Using Modified Whale Optimization And Adversarial Networks Based Classifier For Age Estimation On Facial Wrinkles
Main Article Content
Abstract
Facial wrinkles are considered to be a facial feature, and appear as people get older. The changes of facial wrinkles depend on the nature of skin and muscle contraction. Therefore, the detection of wrinkles plays an important role in applications that depend on facial skin changes, such as face age estimation . While existing wrinkle detection algorithms focus on forehead horizontal lines, it is necessary to develop new methods to detect all wrinkles (vertical and horizontal) on whole face. However, achieving high prediction accuracy without human intervention (such as preprocessing and hand-crafted feature extraction) is currently and potentially a challenge. To solve this problem, this research work proposed a robust feature selection using modified whale optimization and deep CNN-based age estimation method that utilized by conditional GAN. In this work, the age estimation system is divided into five distinct stages. The first step of the proposed system is preprocessing performed by rotation angle. The second step is to extract local features including Gabor wavelets (GW), Local Binary Pattern (LBP), Local Phase Quantization (LPQ), and Histogram of Oriented Gradients (HOG). These attributes are then combined in the third step as a feature fusion method, which combines four different feature extraction methods. In the following fourth step, Modified Whale Optimization (MWO) as a meta-heuristic optimization algorithm is used to decrease the size of attributes and find optimal features. Finally, propose a classification and regression methods to estimate the age and age groups. Initially, GAN is used to determine the age groups, and then CNN used to estimate the ages within those groups. In an experiment using two open databases (the FG-NET database and the MORPH database), when the proposed method in this study was applied to state-of-the art methods for age estimation, the age estimation achieved a higher accuracy than when using low-resolution images.