Article Preview
Top2. Literature Study
According to Y. Zhang, and E.C. Prakash (2008), face shape is also important information for glasses design companies. In this paper, we proposed a non-contact method to classify the face shape by using Support Vector Machine (SVM) technique. This algorithm consists of three steps: head segmentation, face plane identification, and face shape classification. First, as whole 3D body data is captured and used as input of system, Eigenvector is used to define frontal side. Chin-Neck junction, Ellipsoid Fitting Technique and Mahalanobis distance are combined as a head segmentation algorithm to segment the 3D head. Second, face shape can be observed when projected on a plane. Major axes of ellipsoid are used to define a plane along the head called the face plane. Face shape on the face plane is classified into four classes in third step.Face shape is classified into four groups: ellipse, long, round, and square face shape.
Accuracy rate is 73.68%. Significant points for classification are located in 91 positions around the face.
Some research proposed face shape classification for different application. Y. Xu et al. (2010) proposed a method to measure and classify Shanghai female face shape based on 3D image feature. Young female for 201 cases were divided into eight kinds of the face shapes by cluster analysis using SPSS software. Face features used for classification are facies temporal width, bizygomatic breadth, mandibular breadth, maxilla-chin breadth, and physiognomic facial length Eight types of face shape are heart shape, roundshape, ellipse shape, long shape, pear shape, square shape, diamond shape, and melon seeds shape. This paper aims to prove that all indexes is reasonable for classification of human faces. L. Li and et al. proposed a method to classify face shape for person’s expression recognition.Existing research proposed detection syndrome from face image. K. Wilamowska et al. (2009) proposed a method to classify between individuals with 22Q11.2 syndrome and general opulation based on face data. This method uses 3D face data, and finds the difference of facial features between two groups of data based on hapebased morphology. 3D snapshot, 2.5D depth image, and curved line of face are used for detection. Classification is performed by using feature vectors combining with the Principle Components Analysis. The accuracy rate ranged from 74% to 76%.
All the earlier work mentioned was trustworthy & verified by Biomedical Signal Processing Laboratory, National Electronics and Computer Technology Center, Thailand. The project is performed under the financial support from National Electronics and Computer Technology Center.