Research Of Face Detection Algorithms Based On Manifold Ranking And Skin Color Segmentation In Color Image

Face detection is the key technology and very important branch of target detection and target tracking. In recent years, face detection, as the basic technology of detection, has been a leading edge and research hotspot and has received increasing attention in the field of computer vision. It is widly used in aviation, transportation,public safety and other aspects. The usage of face detection can make up for the manpower shortage, it can not only save manpower, material and financial investment,but also improve monitoring accuracy and alarm about abnormal situations to avoid unnecessary loss.Visual saliency is the result of different areas subjectively captured in the image,which makes the certain target in the scene more significant than other areas. And as a result, it can cause our interest and make us exclude other areas in which we have no interest in.In this paper, the visual saliency of the scene image is studied deeply, and the results are applied to the face detection in color images. Through the research about the two saliency detection methods, data-driven bottom-up method and task-driven top-down method, we find that, bottom visual stimulation can better attract the attention of the allocation of resources, and improve the detection efficiency when compared to the guidance detection result of top visual perception task. In this paper,we use the data-driven bottom-up method. This method can detect salient target well in simple background, and also behaves well in running speed and detection accuracy.But when edges of the salient target in the image are similar to the background, this method is not able to extract the salient target accurately and completely, hindering further advanced image processing.The highlights of the main work are as follows,1. Introduce an algorithm combining graph-based manifold ranking with saliencydetection. And the data-driven bottom-up method for saliency detection is adopted in this paper. Through this pedestrian detection processing that meet the characteristics of human visual perception, pedestrian information in the scene will be highlighted,while those background information that we have no interest in will be suppressed. So that salient target can be detected achieving the purpose of pedestrian detection.2. Introduce a kind of image enhancement technology to pre-process image information. In this way, the contrast between target and background areas is significantly enhanced, so that the target can significantly be detected more quickly,which can improve the speed and accuracy of the algorithm execution.3. For the binary image of skin color region obtained through skin color segmentation, this paper introduces the usage of mathematical morphology and connective area detection to capture the most discriminative feature, like edge and connective area, making the detection result more accurate.