Evaluation Method For Stylized Rendering Of Faces

NPR is a rendering technique which is rapidly developed in recent years.It is mainly using some artistic style to draw objects. NPR usesparticular artistic effect to illustrate the scene by different levels of abstractions, which can express the characteristics of scenes and objectsclearly. This rendering technology provides an efficient way to express scenes information, which makes up the limitations of photorealistic rendering methods. Currently lots of non-photorealistic rendering technologies are proposed, and how to evaluate the stylized drawings, including validity of the results, availability, serviceability, etc, is one of the hot issues in NPR.The current NPR evaluation methods are mainly in two ways:one is evaluated based on the exhibition motivation or application area, such asperception of NPR textures, visual effects of illustrations, perceptionof NPR in virtual reality or augmented reality, etc; Another way is by comparing the NPR rendering results with the hand-drawn images by artist. Both evaluation methods need to make sure the purpose of NPR, then evaluate whether the results can achieve this goal. In these methods, hand-drawings by the artist are often chosen as the benchmarksin evaluation.Few of them consider the cognitive mechanism to evaluate and analyze the expression for NPR.Face is an object with a clear field of knowledge.Many methods are proposed for stylized representation of face, such as line drawing, cartoon drawing, portraitdrawing, etc. These methods focus on abstract or process of human faces, but not the true representation of the original face.On the one hand, stylized human face should be with similarity of the original face, on the other hand, the stylized faces should with some artist style, such as pencil sketch style, painting style, and so on. The schemeof evaluate the face stylized representation haven’t been established.This paper proposes anobjectiveevaluation method bases on the normal information and a subjective evaluation method by face recognition theory. Both methods are based on the face domain knowledge.This paper presents the results of a study in which type of line drawings intended to convey 3D faces by usingthe normal estimation method based on the face domain knowledge, including facial expressions, facial orientation andthe regions of five senses. Estimated lines include occluding contours, suggestive contours, ridges and valleys, apparent ridges, and lines from shaded images. Our findings suggest that different style of line drawing has its own advantage in depicting different parts of human face, and that errors in depiction are localized in particular regions of human face. Our study also shows the result that which kind of lines can do better in depicting special parts of human face for different expressions, like eyes, nose, mouth and so on. Finally, we generate optimized line drawing results fordifferent face models to illustrate the efficiency of our evaluationresult.This paper also presents a method to evaluate the subjective recognition of facial expressions, orientations and features, applying to face recognition cognitive mechanism. The method combines face domain knowledge and face recognition cognitive mechanism, and aims at facial line drawings of art and some typical expression patterns of feature lines in 3D face model. This method is to evaluate the expression abilities of different lines in various facial expressions and facial regions by assessing statistic on accuracy rate and efficiency of different stylized lines in recognition process. The evaluation result is used to integrate different kinds of lines, according to evaluation results, to generate optimized facial line drawings. The final validated results show that optimized facial line drawings are more conducive to subjective recognition of face.The objective evaluation methodproposed in this paper helps to understand the object shape effectively. The subjective evaluation method can generate stylized drawings whichare beneficial to face recognition. These methods can be widely used in various stylizations, such as face portrait,oil painting and other artistic expressions.