| Summary: | 碩士 === 國立臺北科技大學 === 資訊工程系研究所 === 104 === In daily life facial landmarks have found wide and increasing applications inclusing face recognition, face tracking, aesthetic medicine, and 3D facial model construction. Realization of these application relies on accurate facial landmark locations. Although automatic feature point extraction techniques such as the Active Shape Model (ASM) and Active Appearance Model (AAM) have been developed, these techniques require training on a pre-defined landmark model and they are mostly applied only to frontal faces. Few and poor results have been reported for automatic locating landmarks in profile faces. This thesis aims to accurately and efficiently locate the landmarks of profile faces without the complicated training process. We also hope to develop an aesthetic medicine software package to help doctors diagnose.
The proposed method extracts the outline of a face image in the sagital view based on the adaptive skin color. Geometric characteristics of the outline contour such as the slope, curvature, or local maximum (minimum), are used to locate the target landmarks including the glabella, nasion, nose tip, subnasale, labrale superius, stomion, labrale inferius, pogonion, menton, and cervical point. By matching the corresponding landmarks in the frontal and profile faces, deviated landmarks positions in the frontal face are corrected. Experimental results show that the proposed method can fast and accurately locate the landmarks of a profile face. The average accuracy of feature point locations is 90.2% under a 6-pixel error tolerance with processing time less than one second.
|
|---|
