A caricature is an extrapolated version of the original face.

  In fact, studies have suggested that people may be better at recognizing a familiar person through a caricature portrait than from a veridical portrait.




   A major contribution of this paper is to define a set of categorical, or nominal, facial attributes. This approach is similar to the attribute and simile features proposed by Kumar
et al. [18], who demonstrated the benefit of this nominal feature representation for recognizing face images. While we present a similar representation, the features proposed here have been carefully defined by a professional artist with experience in drawing caricatures.

   A caricaturist can be likened to a “filter” that only retains useful information in a face for identification. As a filter, the artist uses his talent to analyze a face, eliminate insignificant facial features, and capture the identity though exaggeration of the prominent features. Most of the caricaturists start with the description of the general shape of the head.
They assemble the eyes, nose, eyebrows, lips, chin and ears with some exaggerations in geometrically correct locations (always maintaining the appropriate ratios amongst them);
finally, they include the hair, moustache and beard (depending on the gender or their presence in the face).

  In this study, following the caricaturists methodology, we define a set of 25 qualitative facial features that are classified into two levels (see Figure3). The first level (Level 1) is defined for the general shapes and sizes of the facial components and the second level (Level 2) is defined for the size and appearance of facial components, as well as ratios amongst the locations of different components (e.g. distance of the mouth from the nose).