Gender Recognition using Central Difference Convolution in AlexNet with Haar Cascades

Gender classification is a crucial aspect of face recognition systems in computer vision today. However, accurately determining gender from facial images can be challenging, particularly when dealing with diverse and uncontrolled image datasets. The paper introduces a modified AlexNet Network [1] which has Central Difference Convolution (CDC) [2] layers instead of convolution layers. The approach trains these networks using the CASIA-WebFace [3] dataset and evaluates their performance on labeled faces in the wild (LFW) [4] dataset and CasiaWebface dataset. The dataset is also cascaded using Haar cascading method and the neural model is trained and tested on it. Cascaded CasiaWebface(5000) dataset got the highest testing accuracy of 96.2%.