제안 방법

• In the future, usage of the proposed method may be extended to non-frontally viewed face images, which would require more variation in the geometrical analysis of the proposed method.
• In many approaches, the values for skin color range were extracted from the experimental statistics or a trained image database. In the proposed method, we used YCbCr color mod이 and only CbCr values are clustered into clusters using K-means clustering with eliminating Y factor not to be affected by illumination. Through the clustering process, every pixel of an image belongs to one of the K clusters.
• In this paper, the main concern is how to separate the face area from a still image with a complex arbitrary background. The proposed method is invariant to the number of faces, color, size, position, and orientation of the faces in a still image and does not need any previously trained image set. The remainder of this paper is organized as follows.
• The proposed method was tested to detect faces on still images regardless of the number of faces, the ori­ entation, the position, skin color, and illumination without any pretraining process or a given face database. The whole process was performed through color clustering, cluster scanning, sub-block processing, face rectangle bounding, and face verification.
• To prove the function of the proposed method, three different PC cameras and one digital camera were used to acquire images under different illuminations and orientations. The test images were variously sized color images including up to three faces with various size and rotation applied. In (Fig.
• The proposed method was tested to detect faces on still images regardless of the number of faces, the ori­ entation, the position, skin color, and illumination without any pretraining process or a given face database. The whole process was performed through color clustering, cluster scanning, sub-block processing, face rectangle bounding, and face verification. The experimental tests showed that the proposed method achieved its goal under various conditions.
• To prove the function of the proposed method, three different PC cameras and one digital camera were used to acquire images under different illuminations and orientations. The test images were variously sized color images including up to three faces with various size and rotation applied.