Enhanced SH-CNN: Gender Prediction Based on Convolutional Neural Network

Gender detection and age estimation have emerged as active and vital research areas, finding widespread applications in diverse fields such as biometrics, social networks, targeted advertising, access control, human-computer interaction, and electronic customer services. As the demand for enhanced recognition and classification rates continues to grow, researchers strive to improve the performance of existing methods. In our previous work, we introduced an approach called SH-CNN, which leverages the Discrete Shearlet Transform (DST) as the initial feature extraction layer and the Deep Convolutional Neural Network (DCNN) as the subsequent automatic feature extraction and classification step. The underlying idea was to extract multiple features, obtained through DST’s various decompositions and orientations, from an image. These extracted features served as inputs to the DCNN, enriching the training process and consequently improving the recognition rate. The experimental results demonstrated that the SH-CNN approach indeed led to a substantial enhancement in performance. In this paper, driven by the desire to further boost the recognition rate of our approach, we have proposed an enhanced version called ESH-CNN (Enhanced SH-CNN). ESH-CNN eliminates the constraint of texture differences by feeding only the coefficients generated by the Shearlet transform into the CNN, without including the corresponding face photo. Preliminary results have shown promising outcomes, with a significant improvement in the recognition rate observed when compared to previous approaches.