Summary: | 碩士 === 義守大學 === 資訊工程學系碩士班 === 95 === One of the most important methods to detect human faces in images and video sequences is skin color detection. The techniques for skin-color detection have been wildly used in our daily life, such as monitoring system, human face recognition system and systems for face-image retrieval. However, the effectiveness for face detection may vary with the environmental conditions, such as lighting intensity, lighting color, shadow and background. In addition, it is also affected by the facial expressions, occlusions and their posies.
In this thesis, we aim on the technique for skin-color correction to automatically select proper target images. In our experiments, we compute the mean absolute difference between source image and target images in our database, and pick up the best-matched target image. Then, the color of the source image is corrected by the selected target image. After this step, we extract the skin-color from the corrected image, and then determine whether there is an eye candidate by computing gradient and gray value information. This helps us to determine human faces in images. Experimental results indicate that the proposed method is able to correct 59.4% images automatically. In conclusions, there are several advantages in our method: (1) to select sample image automatically, (2) human race independent and (3) eyeglass wearing independent.
|