Visible to Band Gender Classification: An Extensive Experimental Evaluation Based on Multi-spectral Imaging
Narayan Vetrekar, Raghavendra Ramachandra, Kiran Bylappa Raja, Sushma Krupa Venkatesh, Rajendra S. Gad, Christoph Busch · 2019
The ability of multi-spectral imaging to acquire spatial and spectral details across electromagnetic spectrum, has gained significant importance to perform under varying illumination conditions. Due to which a maximum preference is shown in the recent times to employ multi-spectral imaging in biometric systems. Although there are substantial studies carried out independently using visible image face database and multi-spectral face database operated in several bands, the visible to band gender classification is still an open problem. In this paper, we present gender classification, thereby training the Support Vector Machine (SVM) classifier model using visible face images and testing independently using individual band images for analysis. We evaluate the proposed approach on 79750 sample images which comprises of multi-spectral face database and newly introduced visible face database to present the importance of this study. The extensive evaluation results in the form of average classification accuracy is presented by using 10 fold cross-validation method. The highest average classification accuracy of 95.96±1.68% demonstrates the applicability of visible to band comparison approach for gender classification.