標題: 動態式區塊色調調整實現臉部真實感強化
Face realism enhancement by adjusting tones of adaptable patches
作者: 張育愷
Chang, Yu-Kai
林奕成
多媒體工程研究所
關鍵字: 真實感;色調調整;圖形切割;Example-based method;graph-cut-based method;realizing face;tone adjustment
公開日期: 2014
摘要: Realizing non-real face images is a challenging problem. Different artists may have different painting styles. When seeing a non-real image, people may enjoy the diverse looks of face features. However, when looking at a real photo, they are sensitive to the face contours and feature details. How to transform sparse stroke of paintings into real-face features makes the problem highly difficult. We propose an example-based method to enhance the realism of painted portraits. We use graph-cut based optimization to synthesize a face from multiple sources and form a piece-up face. Next, a quadratic optimization is applied to close the color differences of adjacent patches. In the end of our system, we seamlessly stitch these patches by a two-level blending process. Our experiments show that the proposed method is able to generate a realistic and novel result.
Realizing non-real face images is a challenging problem. Different artists may have different painting styles. When seeing a non-real image, people may enjoy the diverse looks of face features. However, when looking at a real photo, they are sensitive to the face contours and feature details. How to transform sparse stroke of paintings into real-face features makes the problem highly difficult. We propose an example-based method to enhance the realism of painted portraits. We use graph-cut based optimization to synthesize a face from multiple sources and form a piece-up face. Next, a quadratic optimization is applied to close the color differences of adjacent patches. In the end of our system, we seamlessly stitch these patches by a two-level blending process. Our experiments show that the proposed method is able to generate a realistic and novel result.
URI: http://140.113.39.130/cdrfb3/record/nctu/#GT070156642
http://hdl.handle.net/11536/76072
Appears in Collections:Thesis