14086 A new multimodal age prediction image analysis method from hands images of different age groups by neural network model

Abstract

Several papers on age prediction using face images have been published while very limited studies on hands aging have been conducted. Hands aging has a great impact on perceived age and is becoming increasingly important to physicians and their patients. To fill this gap, a novel age prediction method based on the hand images is developed. Age prediction models based on face images are well investigated mostly because of the persistence of massive datasets like IMDB-WIKI containing more than 500,000 face images. We captured face and hands images of 994 Indian women with age ranging from 20 to 60 years with different skin tones by VISIA CR. We developed two deep neural network architectures for the face and back of the hands, that can be used separately or in conjunction for age prediction. A novel neural network architecture works on hand images and predicts age based on the seven specified areas of the hand image such as central and bottom part of the backhand, knuckles areas of fingers. Predictions for all these seven areas are aggregated for the final age determination. This method based only on hands images showed a great association between actual age and predicted age with mean average error (MAE) of 5.89 years. We got MAE of 6.8 years for face-based age predictor. Our research can help to measure the complex effects of anti-aging products or any dermatologic procedures after treating aging hands objectively and as part of a multimodal age predictor.