Semi-supervised Clusteirng Algorithms In Face Detection Application

Face Detection technology attracts people’s more and more attention in thisdaily lives. Especially in the field of scientific research and national defense, facedetection is growing at a rocket speed of development. For this technology, it is ahigh-tech composite technology which combines multi-disciplinary, multi-field andmulti-level knowledge. Face Detection not only includes theory of biology,physiology, mathematics, computer science, but requires researchers to continuouslyexplore the variety subtle Facial changes and influence of surrounding environment.Therefore, it is an difficult, interested and challengeable research topic.The clustering method has always been a very crucial technology in the field ofdata analysis, data mining and image processing. It can help researchers to conductstatistical analysis on the data of unknown classification. Traditional clusteringmethods don’t display satisfied clustering effect. Most of them are due to lack ofinformation of data sample itself as a guide. So, in recent years, many experts beganto learn strategies of semi-supervised in the field of machine learning. And then usethis learning method to the cluster direction getting an amazing result.In this paper, learning strategy based on SKDK is applied to the color clustering.In this clustering process, First of all, the small marked samples are used to buildSeed set and guide the color clustering. What’s more, build the skin color modelaccording to the probability of each pixel cluster statistics. On this basis, depends ontraditional knowledge of mathematical morphology to process image in order toeliminate noise in the image. Then use these face candidates as later classifier inputdata. At last, depend on some outstanding classification to finish Face Detection.Iamge is through roughing of color clustering, which rule out most of irrelevantbackground. Under this circumstance, the subsequent face detection speed andaccuracy can significantly be improved. At subsequent section, this paper proposes aface detection algorithm based on improved ICA, which improve and accelerate theconvergence speed to traditional Newton iterative algorithm. But this algorithm inthe scenario of multiple face detection doesn’t show a good ability. After this, thispaper use Real AdaBoost algorithm with candidates face to complete the facedetection in a variety of complex scenarios. Experiment of result obtains a goodeffect.