Improving Gender Classification Accuracy in the Wild

Abstract

Abstract. In this paper, we focus on gender recognition in challenging large scale scenarios. Firstly, we review the literature results achieved for the problem in large datasets, and select the currently hardest dataset: The Images of Groups. Secondly, we study the extraction of features from the face and its local context to improve the recognition accuracy. Different descriptors, resolutions and classifiers are studied, overcoming previous literature results, reaching an accuracy of 89.8%.

Fusion of Holistic and Part Based Features for Gender Classification in the Wild

“…The reader must observe that these results were achieved not following a lights-out, black-box testing methodology. Focusing on GROUPS, with the exception of the protocol described by Dago et al [9], used in [4] too, the adopted protocols are not easily reproducible. The fact that GROUPS is currently the most challenging in the wild dataset, has convinced us to focus on this dataset.…”

“…Two recent results support the approach described in this paper. On the one hand, the extraction of features at different scales may benefit the GC performance [2,4]. In [4] the features are extracted from the face and its local context, thus, the face is analyzed at different resolutions.…”

On using periocular biometric for gender classification in the wild

“…They are analyzed in the range [1,8] and [1,6], i.e. analyzing grid from 1 × 1 to 8 × 6 cells, making a total of 6 × 48 = 288 variants.…”

“…For this problem, this effect is added to the proven accuracy improvement as claimed in [8] and the reduction of ambiguous cases occurrences in [10].…”

“…This approach is designed based on our previous proposals that combines facial and head and shoulders features for this task, see [8] and [9]. However, those systems focused on the complete facial and its local context, i.e.…”