Skin balancing

Abstract

Portrait photos have a relatively profound meaning compared with landscape or object photos. Since camera is not that smart as our eyes, reproduced color of portrait differs from scene illuminants. When multiple portraits gathered together for a distinct storyline conveyance, the individual difference in color become obvious and could lead to content bias unintentionally. Furthermore, instead of computing methods, appear of portrait photos in professional level relies more on expert’s adjustment in software slide by slide. In these regards, this research aims to understand how people perceive multi-illuminant portraits on one scene, and propose computational method to standardize the image quality for portraits based on the understanding. To be specific, a user study was firstly conducted to investigate how people respond to portraits under warm, neutral and cool color tones, simulating multiple-illuminants. Evaluation results and outcomes of creative tasks were analyzed both quantitatively to supplement the findings from qualitative analysis. Based on the findings from responses of the end users, techniques of portrait enhancement were reviewed and selectively adopted in order to develop method to enhance color for these media contents. We proposed a method as ‘Skin Balancing’ to conduct absolute color adjustment based on target skin color for portraits. Four target skin colors were provided to reproduce portraits with a standard image quality with minimal content bias. User tests revealed that the proposed method provides satisfactory outcomes in comparison with original image or existing methods. Lastly, the practical values and possible applications of the Skin Balancing were widely explored while interviewing experts from diverse domains together with reviewing the limitations that current research has.