Validating Machine Vision Competency Against Human Vision

Abstract

This work elucidates the human intelligence performance and machine intelligence in geographical region-wise face classification incorporating a sample of 120 human identifiers and computational models like convolutional neural network and colour local binary pattern. A novel Indian colour face database is created consisting of 2010 distinctive face images of east and south regions. On human side, an automated human intelligence system is established to evaluate the visual capabilities of human. On machine side, the authors trained two CovNets, one comprising more layers trained with 1800 normal face database images and another one trained with 1000 contoured images of face obtained by canny edge detection approximation method, to estimate the human intelligence response that found face shape the more discriminative feature among other face features. Experimental results showed the human classification proficiency (96%) stood superior to the machine algorithms even in challenging aspects.