Hybrid Approach for Synthesis and Classification of Facial Images for Age Determination

Abstract

Human Age Estimation from facial images has been an active research topic in computer newlinevision due to its diverse field of application such as security and surveillance, biometric, newlineentertainment, Internet access control, Electronic customer relationship management and newlineInformation security. Human face renders both local and global facial features where newlineglobal refers to facial appearance and local represents the wrinkles. The major newlinecontribution of the work is hybrid approach for age classification and estimation using newlineboth local and global features identified and extracted using face images. A novel method newlinefor feature extraction is implemented for shape and texture in face image. newlineIt was proposed to deliver a method that exhibits facial aging in humans. In the first newlinestage, a model is developed that express a transformation of physically based newlinecharacteristics that has refined facial deformations that undergo with age. This model newlineimplicitly shows the physical and geometric variations in individual facial characteristics. newlineIn the second stage, a process for texture transformations was modeled that exhibits facial newlinewrinkles that are often visible during the aging phenomenon. For this transformation newlinemodel as input, FGNET (Aging Database) is used that consist of face images with age newlineinformation of different subjects at their different ages. The training for the extracted newlinefeature dataset is performed for both global and local features individually as well as newlinecombined using various classifiers and found results to be significant compared to results newlineof other researchers. To further enhance the estimation capability the ensemble learning newlinetechniques were implemented that improved the overall efficiency of the proposed model. newlineThe improved efficiency is 95% as compared to efficiency of 89.13% achieved by Abbas newlineet al.(2018). newlineFurther, the global and local datasets are fused to which deep learning techniques were newlineapplied for age classification and estimation mechanisms to further improve the newlineefficiency of the proposed system. The Deep