Age Estimation From Face and Hand Images Using Convolutional Neural Network

Abstract

These days, age estimation is attracting a lot of attention. It has been a mystery long since researchers started working on it. They are still trying to find out what genetical and biological factors are responsible for it. Facial images are widely used to estimate age of a person. Since facial image contains a huge number of biological characteristics, it can say a lot about a person’s age. But we can’t pinpoint what indicates this age. Lots of factors are playing important role here, such as Environment, habit, mental pressure, biological age and so on. Moreover, a male person may grow beard and uses antiaging cosmetics and other chemicals. A female may use antiaging cosmetics, makeups and other chemicals which effect their facial skin. Some research indicates, rear of a person’s hand contains important biological data as well. Ridge or crease of hand may contain biological markers of a person’s age, which normally do not get affected by cosmetics and chemicals, though environmental affect is impossible to avoid. In this research, we tried to find out a way of age estimation from face and hand images using convolutional neural network (CNN). Though there are lots of age estimation methods available which used only facial images, we tried to do it better by including hand images. We have worked on a hypothesis that hand images may provide important details to this research area and will improve efficiency of age estimation process. When it comes to process image features, CNN have an incomparable advantage. CNN has proven to be much more accurate at estimating age of facial images than traditional methods. We have applied same algorithms and codes on hand images, but could not get effective result as we expected. Since there are no works on age estimation from hand images available, there is no way to implement and try any algorithms specifically designed for this purpose. Thus, we had to do it from the scratch. We could not get expected result, but we think this research may provide a ground to the future works on this field.