EGA construction The EGA Face Database
An increasing number of research efforts presently focus on assessing the hypothesis that a preliminary face categorization can be used to contain the search space during identification operations, and support an improvement of recognition performance. The underlying assumption is that a sample image is only matched with those pertaining to the same category. The interest for related investigations stems from the consideration that, despite recent advances, presently available systems are not yet accurate or robust enough to target under-controlled yet high security environments. The exploited categories are mostly related to soft-biometrics, such as gender, age, ethnicity, or to their combination as well. Experimental results demonstrate a general improvement in recognition accuracy, while reducing operation time, in applications such as multimodal recognition, face age estimation, and face re-identification. However, available datasets are not organized according to any categorization, and it often happens that either ethnicity or gender, or age, are not uniformly represented. Many datasets are collected by enrolling students, therefore the prevailing age range is 20-35. Moreover, the geographical location of enrollment operations often influences the prevailing ethnical composition of the dataset. On one hand, this may make experiment set-up quite laborious, with preliminary hand selection of appropriate subsets of images. On the other hand, results may be biased by the lack of a face dataset sufficiently large and fairly representative. We provide here an automatic procedure to build EGA (Ethnicity, Gender and Age) face database, a larger multi-racial database, designed and implemented to overcome these limitations.
- Next >>