ROI sensitive analysis for real time gender classification
Marcos Rodrigues, Mariza Kormann, Péter Tomek · SHURA (Sheffield Hallam University Research Archive) (Sheffield Hallam University) · 2014
This paper addresses the issue of real time gender classification through texture analysis. The purpose is to perform sensitivity analysis over a number of ROI-Regions of Interest defined over face images. The determination of the smaller ROI yielding robust classification results will be used for fast computation of texture parameters allowing gender classification to operate in real-time. Results demonstrate that the ROI comprising the front and the region of the eyes is the most reliable achieving classification accuracy of 88% for both male and female subjects using raw data and non-optimised extraction and classification algorithms. This is a significant result that will drive future research on optimisation of texture extraction and linear discriminant algorithms.