If you Googled ‘Face shape spectacles’ or similar search terms, you’d get plenty of URL references pointing to a variety of optician homepages that provide free advice for best aesthetic effects upon their customers wearing designs of their frames. To classify human face morphology most (beauty) experts gravitate towards seven dominant face shapes: oval, square, round, oblong (aka rectangle), diamond, heart (aka inverted triangle), and triangle. Tipheret proposed objective criteria to efficiently perform this classification. However, the face shapes of many real persons often fall between discreet classes, and it is quite a challenge for a computerized system to accurately predict without human intervention which shape class a given face belongs to. Nevertheless, IBV has researched the possibility to automatically identify face shapes from end-user portrait images, based on the automatic positioning of critical reference points, by measuring relevant distances among those points, and finally applying purpose-made prediction algorithms (Discriminant Analysis and Fuzzy inference - associative matrices- techniques were used to this end). Results of this work and conclusions thus far are also presented in this article. The effort was guided by the goal to create a sales environment for the selection and purchase of pairs of spectacles, personalized to end-users’s head and face characteristics for optimum comfort and aesthetic result (persona).
This is an extract of an article I wrote about automatic frame selection. Link to the original post and download the remaining article, in support of our project.