CAJ | 학술논문

为了克服传统灰度积分投影方法无法有效定位旋转人脸图像中人眼的缺点，提出了一种基于极坐标系的灰度积分投影方法。利用肤色特征对给定图像进行人脸区域的确定，在人脸区域内按极角方向进行灰度积分投影，确定出人眼所在角度，将人眼角度方向的像素灰度值做水平方向积分投影，从而确定出人眼的位置。该方法能够实现同一幅图像中多个不同姿态人脸的人眼定位。大量的仿真实验表明，该方法的识别性能对人脸的旋转变化具有良好的鲁棒性，能够提高灰度积分投影方法对人眼定位的适用范围。
위료극복전통회도적분투영방법무법유효정위선전인검도상중인안적결점，제출료일충기우겁좌표계적회도적분투영방법。이용부색특정대급정도상진행인검구역적학정，재인검구역내안겁각방향진행회도적분투영，학정출인안소재각도，장인안각도방향적상소회도치주수평방향적분투영，종이학정출인안적위치。해방법능구실현동일폭도상중다개불동자태인검적인안정위。대량적방진실험표명，해방법적식별성능대인검적선전변화구유량호적로봉성，능구제고회도적분투영방법대인안정위적괄용범위。
Conventional gray-level integration projection can not detect the eyes in the rotating face image. In order to overcome this shortcoming, an improved gray-level integration projection method in the polar coordinates is proposed. Face regions are extracted from the image by the feature of skin color. Gray-level integration is projected to the polar angle in the face region to detect the angles of eyes. Pixels in the direction of eyes are projected to the horizontal direction to deter-mine the position of eyes. The method can detect eyes from several different posture faces in the same image. A lot of sim-ulation results show that the detection performance of the method has good robustness to the rotational variation of face. Thus, the method can expand the eye detection application of gray-level integration projection method.