Edited Conference Papers

We propose an age recognition method that compares a pair of image instances from the same gender for the three conditions of reflexivity, antisymmetry, and transitivity corresponding to older, similar, and younger than respectively. The approach uses a small number of reference set to address performance issues due to the lack of datasets for age recognition. Experimental results demonstrate that the use of a few optimally selected class references for the comparison resulted in a dramatic improvement on the baseline. Thus, with a Mean Absolute Error (MAE) of 4.07 and Cumulative score (CS) of 93.87% on the adience dataset, our proposal shows superiority among other gender-inspired age recognition proposals.