Research On Feature Extraction And Classification Algorithm Of Face Image

Pattern Recognition Technique(PRT) has been a very popular subject since the 1960s century,whose theries and methods are also very important in many other subjects and fields, promote the development of artificial intelligence,let human have come into an intelligence time.Pattern Recognition problems are divided into two independent parts:feature extraction and discriminant.The paper will explore both such two problems.Feature extraction methods are emphasises in the paper,therefore,the extention work of classifiers will been expatiated first.As we know,the human face is color image,and our eyes can distingguish several thousand colors[1],while only 10¹5 level grayer.Therefore,color facial images include more disciminant information.If we can use the color information well besides the grayer information reflecting image's shape and configuration,we will get more disciminant information to improve the disciminant rate[2].While most of the classical pattern recognition methods are based on grayer images[3],transforming color images into grayer images before discriminant steps,which lose some feature information of color images and has resulted in classical classifiers[4],including the Nearest Feature Classifiers,are used in grayer images discriminant only,while the discriminant capability in color images are hardly validated,which will been completed in this paper,we extent the Nearest Feature Classifiers to color face recognition,and get very good experimental effect on ar_color color face database,which shows that such classifiers are effective in color face recognition and the necessary of our work in this paper.Cost-sensitive learning methods have been used in pattern recognition,but hardly in feature extraction step,including Jiwen Lu's work[5],which aims to get the least error cost but higher disciminant rate.Therefore,if cost-sensitive can been used in feature extraction step and get good discriminant features,there are not such answers.Therefore,the second work in this paper extent the cost-sensitive to feature extraction step and get a series of fine discriminant features on three database(ar₆0 facial database,palmdata database, Concordia University CENPARMI database)for the first time, therefore,only if we have a good project,the cost-sensitive learning method will have a bright future in improving discriminant rate.