Article Preview
Top1. Introduction
Face images convey various information about the emotional state, identity, gender, ethnic origin, head orientation, and age of the person. This information is very important during face-to-face communication among humans. The facial information during interaction is made possible by humans to identify and interpret facial gestures and faces accurately in areal time manner (Lanitis, et al., 2004). Here, the human ages are predicted from the related facial images (Thomas and Rangachar, 2019)is very challenging in the computer vision. The popularity of this topic is due to their potential applications, like effective filtering in criminal investigation human-computer interaction, intelligent advertising, and so on (Dotu, et al., 2011)(Hakeem, et al., 2012)(Xie and Pun, 2015). The major challenge is owed partially to uncertainty in the face aging processes that manifest as the personalized conditions are under the same age. In this case, the uncertainty is caused by extrinsic and intrinsic factors, respectively (Xie and Pun, 2015).
The estimation of human age is very significant in computer vision as well as image. Age estimation has been played various applications, such as audience-measurement systems, Age-Specific Human-Computer Interaction (ASHCI), security surveillance, and the soft biometrics. In addition, the age estimation is very useful in various face-related tasks, including face recognition and detection (Luu, et al., 2009)(Dong, et al., 2016). The major issue faced by age estimation is due to the ageing effects with other facial variations. Personalized ageing patterns, and the uncontrollable nature of ageing process, and at last, lacks of representative and comprehensive age face images dataset to train the precise model are some complexes in facial age (Han, et al., 2013)(Bekhouche, et al., 2016). The difficulties caused by age estimation from the facial images are constructing complex feature descriptor techniques, which are tried to mimic animal behaviour, or the fine-grained facial regions for performing accurate alignment based on multiple facial fiducial points. Thus, the feature extraction approach is very difficult for reusing, and in several cases, it takes more time for extraction (Taheri & Toygar, 2019).
Several kinds of research based on age estimation are performed for extracting aging features from the images. Some of the examples for feature extraction are: age manifold, Active Appearance Model (AAM) (Cootes, et al., 2001), Aging pattern Subspace (AGES), and biologically inspired features (BIF). In addition, the image-enabled age prediction approaches consider the face image as the texture pattern for predicting demographic attributes (gender, age, and the ethnicity) histograms of oriented gradients (HOG), local binary patterns (LBPs) (Al-Shannaq and Elrefaei, 2019), Binarized Statistical Image Features (BSIF), BIF, and the Local-Phase Quantization (LPQs).Consequently, the multiclass classification scheme considers the age value as a separate category, and then the classifier is used for classifying the age. Hence, standard classic methods, which include multilevel perception, AdaBoost, K-NN, and Support Vector Machine (SVM)(Beno, et al., 2014),are utilized for performing accurate age grouping and prediction. The regression method maps the feature space into the age space to obtain optimal regression function, and the nonlinear regression methods, such as Support Vector Regression (SVR)(Ghuge, et al., 2018),Gaussian Process(Ratre and Pankajakshan, 2018), and the quadratic regression for solving age estimation issues. However, the regression method considers various age categories as increasing linear relationships and failed to reflect the diversity of ageing process (Liao, et al., 2018).
The main aim of the age estimation strategy is to identify the person’s age using the image. Age detection has been paid great attention in several applications and is suitable for identifying the features that represent the age of face image. At first, the input image is fed to the pre-processing step, which is performed using the Viola Jones algorithm. After pre-processing, the feature extraction is carried out based on LTDP, AAM, and the novel feature extraction approach Deep appearance model. After that, the classification is done based on the extracted features using the proposed CSDBN. The proposed CSDBN is designed by integrating CSA and SCA. Thus, the CSDBN classifier finds the age of face image.
The main contribution of the research paper are enlisted below,