A Study Of Gender Recognition And Age Estimation Based On Face Image

Facial image contains rich information,from where many useful messages can be extracted,such as identity,gender,age,feeling,ethnical features,physical condition,and so on.Presently,the research on facial detection and recognition is considerably developed,and is extensively applied in intelligent cameras and access control systems.However,the interpretation of the messages from the facial detection and recognition is far from being mature.Gender recognition and age estimation based on facial image is the extending application of facial image.Because of the potential applications in the aspects of identity recognition,man-machine interaction,video retrieval and robot vision,gender recognition and age estimation has become a hot research point in the domain of computer vision and mode recognition.On the basis of predecessors' research,and after a systematic research on and further exploitation of gender recognition and age estimation based on facial image,this essay has tried some improvement method,and come out with some significant results.The main content is as follows:1.Sample preprocessing.There can be various situations with the image from the face database,such as non-facial background,different sizes of facial picture,uneven brightness and rotation,etc.Therefore,the picture of the facial image area should be extracted for preprocessing and normalization before gender recognition and age estimation.In this essay,Adaboost algorithm and Haar feature of the facial image are used for automatic facial detection.Geometric normalization of the facial image is achieved via coordinate locations of the two eyes.That can reduce the influence of the location message and background noise on the following experiment.2.Experiment is performed for facial image gender recognition on CAS-PEAL face database.LBP Feature,HOG Feature and Gabor Feature are extracted respectively as gender feature,and SVM is used for gender judgment.The experiment shows that all the three methods are applicable for gender judgment,while HOG Feature method comes out the best,with a recognition rate of92.13%.3.The essay discusses the two relevant parameters when extracting HOG Feature,analyzes andsummarizes the best method for selecting HOG operator parameters in facial image analysis,and offers the recommended value force HOG parameter selection.4.In this essay,different ages are classified into a couple of age groups,therefore,the age estimation problem is transformed into multi-class classification problem.The experiment of face image age estimation is performed on FG-NET face database.Since the age change on face is a complicated procedure,multi-feature fusion is applied in the essay.The quantity of textural feature extracted with LBP Feature method is rather limited,while HOG Feature can well describe the message of image edge and direction with the distribution of pixel gradient magnitude and gradient direction,therefore,LBP Feature and HOG Feature are chosen for fusion in the essay,and three different methods for feature fusion are discussed.It is shown in the experiment that the recognition rate of fused features is higher than that of a single feature,among which LBP+HOG fused feature method has the best assessment result,with the total assessment precision of 78.00%,which is at least 4.00% higher than that of single feature recognition.5.Consideration in this essay is given to both feature-level fusion and decision-level fusion.This method,which is based on multi-feature fusion and multi-classification classifier for facial image age assessment,is offered: LBP Feature,HOG Feature and LBP+HOG fused feature are applied for training of three classifier sets.The two sets of single feature classifiers are first used for classification.When the two sets produce different results,the classifier trained with fused feature will be called for classification,and the result produced is considered as final.In this method,the strong classification capability of the fused feature classifier is used to solve the problem of incorrect classification by single feature classifiers.It is shown in the experiment that the total recognition rate of the method discussed in the essay is 78.67%,which is based on multi-feature fusion and multi-classification classifier for facial image age assessment,0.67% higher than that of multi-feature method.That proves this fusion method can reduce some unnecessary steps,while the total precision of age assessment is raised.