I'm working on a project that involves analysing an image of a 3D avatar face and determining how realistic it looks. One of the metrics for this realism is based on the skin, more specifically its texture.
I'm using the color FERET database to compute averages that correspond to a human face, however I'm unsure of what technique I should be using to analyse a skin ROI of a human face i.e. the cheek, which I would compare to a similar ROI from a 3D avatar.
I understand this question is rather complicated, since there is no objective measure for realism and thus no exact method may exist to make this comparison.
Currently I'm using LBP and comparing the histograms of the human ROI and the avatar ROI with Chi-Squared distance, however the differences of comparing a human and an avatar versus a human and a human are so minute that theres almost no point.
Is texture the best way to approach this, if so is LBP & histograms comparison the most appropriate method?