Three Fold Classification Using Shift Invariant Deep Neural Network

Abstract

Due to increase in digitalization in recent years, there is huge amount of images are available with different parameters like pose, scale, lighting, and occlusion. Predicting their age, gender, as well as race is a resilient task. In past year CNN based model plays a virtuous role in prediction of age and gender. In this research we proposed a threefold classifier using shift invariant Deep Neural Network for age, gender and race classification in a same model. It will contain multiple layers like maxpooling, covlutional block. For training and testing of our model we have used UTKFace dataset that contain 23705 images with varying parameters. For age estimation, gender and race classification we have used neurons at the output layer. Output Neuron are taken 1, 2 & 5 for age, gender and race classification respectively. Loss function relu is used for age estimation, for gender and race classification sigmoid function is used. In this study we convey that a considerable upgrade in age estimation, gender and race classification be obtained by using shift invariant Deep Neural Network. The obtained result for age, gender and race are 0.08, 87.66% and 76.22% respectively. Results convey that suggested approach provide a great performance than the other modern model.