Study Of Face Detection And Localization Based On Skin Color Segmentation

The face detection is to locate the faces and determine their size, position andamount in an image. The application of face detection has been far beyond thecategory of face recognition in the aspects of retrieval based on the content, digitalvideo processing and visual inspection. To sum up, the face detection algorithms canbe divided into the method based on skin color detection, the method based on theshape detection and the method based on features detection.The paper researches the threshold segmentation technique first and puts forwardan improved two-dimensional threshold segmentation algorithm based on entropy andan improved two-dimensional threshold segmentation algorithm based on maximuminterclass variance, and the experiment results are given in the end. Then the paperresearches face detection and location algorithms and the improved algorithms basedon the technology for the single front face images with complicated background, andproposes the improved light compensation algorithm, the improved colorsegmentation algorithm based on the maximum two-dimensional entropy, theimproved color segmentation algorithm based on the2D OSTU, the colorsegmentation algorithm based on YIQ color space and the2D OSTU, the colorsegmentation algorithm based on integral image and particle swarm optimization andthe color segmentation algorithm based on double color space, and verifies thevalidity of all algorithms in the environment of MATLAB2010. Finally, on thefoundation of skin segmentation, this paper proposes a face location algorithm basedon facial rectangular characteristics and a face location algorithm based on improvedprojection method, and a simple face detection and location system is designed in theplatform of Visual Studio2008with VC++programming language.The improved color segmentation algorithm based on the maximumtwo-dimensional entropy and the improved color segmentation algorithm based on the2D OSTU and the color segmentation algorithm based on integral image and particleswarm optimization proposed in this paper improve the traditional algorithms at theaspects of how to reduce the amount of calculation and how to speed up the testing,and the experimental results also verify the accuracy and feasibility of thesealgorithms. The color segmentation algorithm based on YIQ color space and the2DOSTU uses the I component with appropriate processing as the skin similarity graph to do skin segmentation, and the experiment result shows that the algorithm canobtain better segmentation effect for most of the test images. The color segmentationalgorithm based on double color space chooses the test color space by comparing theaverage of the channels of R, G and B with the average of the channel of B, in orderto improve the algorithm’s robustness, and the experiment verifies the validity of thealgorithm too. The face location algorithm based on facial rectangular characteristicsuses the area of the skin region which has been processed by median filtering with anappropriate template and the ratio of facial length and width to construct optimal faceregion rectangle for face localization. The face location algorithm based on improvedprojection method uses morphology knowledge to automatically detect the connectedregion, and eliminates the false skin area according to the area of the connecteddomain, and then uses the traditional projection method to find the boundary of theface region. The experiment results show that these algorithms can detect and locatethe face in an image effectively.