Research On Face Detection Algorithm Under Complex Light

Face detection has been widely used in applications,such as video storage,fatigue driving,human-computer interaction and so on.How to improve the face detection rate under complex lighting conditions is an important research direction because that complex illumination has a great influence on the rate of face detection.This paper mainly focuses on the issue of face detection under complex lighting conditions.The content is as follows:(1)The enhanced model of multi-filter fusion we proposed in this paper aimed at solving the problem that complex lighting environment hides image information.Firstly,the model uses multi-scale Retinex algorithm which remove light components and obtain the reflected components.Secondly,guided filtering be used in this model to get detail layer,which to make the high-frequency enhancement.Experimental results show that the enhanced image not only eliminates the effect of light on the image but also enhances the details of the image effectively.(2)Global low-rank decomposition is applied in this model,which could solve the problem of noise increase caused by Retinex algorithm which stretches the dark area of the image.Experimental results show that the noise can be removed effectively.(3)HLBP is used to described the face feature,which is designed to improve the robustness of face detection algorithm to lighting.HLBP's performance is better than the Haar feature significantly,not only because of rotation invariance and gray invariance,but also due to not sensitive to light.Experimental results show that the HLBP feature is sensitive to the edge and texture of the images.Under the normal lighting conditions,face detection rate of extracting HLBP feature increased by 3.07%compared to Haar feature.Under complex lighting conditions,this paper uses different illumination angles such as(0°?12°,(13°?25°),(26°?50°),(51°?77°),and(78°or more)to do some experiments,and results show that the face detection based on enhanced model we proposed has a higher face detection rate compared with the traditional algorithm.