Feature Clustering and Synthesis of Caricatures

• Main Authors: jr-cheng dung, 董致成
• Other Authors: 白敦文
• Format: Others
• Language: zh-TW
• Published: 2004
• Online Access: http://ndltd.ncl.edu.tw/handle/88401668533673615921

碩士 === 國立臺灣海洋大學 === 資訊工程學系 === 92 === In this thesis, we propose shape clustering and classifying algorithms for facial feature vector based 2D cartoon caricature generating systems. The digitized images are imported into the proposed system and their respectively defined facial feature control points are observed automatically or adjusted manually by users. This feature controlling points are transformed into vectors and clustered by proposed machine learning algorithms based on Fuzzy C-mean technologies. After feature training, the systems provide different groups of representative shapes with respective to each facial organ, such as face contour, eyebrows, eyes, nose, and lips. Therefore, by training grouped feature vectors, efficient and effective redrawing techniques are then able to be applied for generating personalized cartoon caricatures. We profoundly believe that appropriated classification techniques for facial features are important research topics for future related multimedia applications.