Age Estimation from Facial Images Using Transfer Learning and k-fold Cross-Validation

Abstract

The increasing use of video-based security systems and robotics has increased research on the image analysis of human faces. Thus, face recognition, face detection, gender classification, and facial expression recognition have attracted much attention in digital image processing field. [1, 2, 3, 4, 5]. Estimating the age of a person from the analysis of his/her face image is a relatively new research topic. This thesis is focused on exploring different CNN approach on increasing accuracy in age classification. To achieve higher accuracy we used a few pre existing models and used pre trained weight which is trained for face detection,on Wild and Youtube face dataset. We also fine tuned the model and used k-fold validation. This method shows us higher accuracy. Then we compared our work with already existed papers. In other words, we tried to increase accuracy in age classification. We also compared our work with pre existed paper to show performance of our model relative to other models