The capsule network has evolved as an alternative to resolve large data requirements of the convolutional neural network amidst their inability to recognize object pose and deformation. Hence, this paper utilized the equivariance property of the capsule network to develop an improved human gender classification system with the capability of classifying human gender using either facial images or fingerprint images. The capsule network architecture was redesigned with five convolution layers using squashing activation layers to efficiently extract global features of fingerprint and facial images from a developed hybrid dataset. To ensure that the system output images are identical to the input images, a computed reconstruction loss weight of 8.192 was used. The obtained results from the evaluation of the developed system show that the system has an accuracy, recall, precision, and false error rate of 98.3%, 96.5%, 98.4%, and 0.136 respectively. The system has a 2.72% improvement over the existing hybrid support vector machine and Inception-V3 gender detection system accuracy. Furthermore, the developed system has the added advantage of classifying any fingerprint images or facial images into their respective gender without any redesign or modification.