기울어진 얼굴 영상에서의 특징 추출 및 계층화에 관한 연구 = A Study of Feature Extraction of Tilted Facial Image and It's Hierarchy

다국어 초록 (Multilingual Abstract)

In this paper, we propose a facial recognition method that accommodates various scale and slope information. Standard facial images and tilted facial images are captured by CCTV camera, transformed to binary image through use of filering and edge detection method. Depending upon the slope information taken from the relative position of left and right eyes, the tilted images are rotated. Then the facial images are subdivided into functional components using the projection profiles. Consequently, positional information represented by a 3 by 3 matrix was used to yield slope and ratio of distance. To further increase the efficiency in recognition, fuzzy linguistic variables are calculated from the real numbers thus produced, and put into an hierarchical order.
Experiments show about 90% of recognition rate for the sample images of 30 frames when the slope of input image was considered, and this result can be a great advance compared with the slope of input image was not considered.