Geometric-Based Facial Feature Localization Method

3894 Words16 Pages
Abstract This paper is a note on a geometric-based facial feature localization method, which is proposed by Dehshibi and Bastanfard called Combined Projection Function (CPF). 1. Introduction This work sorts out the problem of facial feature localization featuring Linear Principal Transform (LPT). Facial feature extraction, in general, refers to detection of the eyes, mouth, nose and other important facial components. It is the most significant, preliminary stage in developing applications like Face Recognition [], Facial Expression [], Face Detection [], Age Classification [], Gender Classification [], etc. Due to the role and importance of the localization stage, various algorithms have been proposed since the last three decades. The proposed methods usually express in combination with other applications and we can firmly say that the number of research, which concentrates on feature localization, is few. Furthermore, specific features such as eyes and ears [] have been considered more thoroughly. The most important factors, which keep this research area as an open problem, are changes in the face perimeter; e.g., change in the location of head, background, illumination, facial expressions, etc. Despite the importance of locating the various face features, there is a general agreement, which states the eye is the most important feature of the face []. This is due to several reasons among which are the following: Existence of eyes verifies that the object of interest is human. Factors which influence the face appearance have less effect on the appearance of eyes. For instance, the eyes are unaffected by the presence of facial hair, and are little altered by small in-depth rotations. Knowledge about the eyes’ position ... ... middle of paper ... ...Moreover, beside the three previously mentioned features, we could also extract top and bottom of the eyebrow line. The result is depicted in Figure 7. 4.2 Locating Nose, Lips and Chin using LPT Although locating the nostrils due to low contrast in this area is fairly hard, LPT has high ability to overcome this complexity. The overall process of localization takes place in two stages. First the nose perimeter is extracted from the face image. In this stage, different methods including anthropometric-based, template-based, and appearance-based can be used to extract the region of interest. Then, LPT is applied to this region. As is evident from Figure 8, the global maximum of the graph shows the tip of the nose location. In the previous work [] this landmark is extracted by applying two types of projection function. Therefore, LPT can reduce the execution time.

More about Geometric-Based Facial Feature Localization Method

Open Document