Facial Gender Recognition Via Low-rank And Collaborative Representation In An Unconstrained Environment

Facial gender recognition is an attractive topic in the field of computer vision and machine learning.But,most of the existing facial gender recognition methods are always suffering from the problem of weak robustness when working on the unconstrained environment.For this reason,the low-rank decomposition and collaborative representation are exploited to raise the precision and the robustness of the facial gender recognition algorithm.Our study use the low-rank decomposition to minimize the negative effect caused by image corruption and various face poses.In the phase of recognition,the collaborative representation mechanism,which substitutes1l-norm regularization with 2l-norm regularization to enhance the discriminant power,decrease the computational cost of the gender recognition system and solve the small-sample-size problem of facial gender recognition.In the experiments,the proposed method is evaluated on the several benchmarks.The result demonstrates that our method is able to achieve better performance than the current state-of-art facial gender recognition approaches.The work of this paper is as follows:(1)This paper detailedly descriptes facial extraction method and classification algorithm of facial gender recognition,and conduct the experiments with classical method of facial gender recognition.The result of the experiments are recorded.Finally,theses system of ficial gender recognition is analyzed detaily according to the performance.(2)Recently the sparse representation based classification(SRC)has been successfully used in face recognition,so it is applied in facial gender recognition in this paper to get the better capability of classification.Then,to obatain the higher accuracy,the RSC is choiced as our classifier.Such a method is more robust to classification of the coverd and occluded facial images.It is because that the normal pixels are allocatted the high weight and the corrupted pixels are allocatted the small weight.This algorithm will have the more performance according to these informations.Through the RSC can improve the discriminative power of the facial gender rencognition system,this algorithm is eliminated when which classifer is used in our system becasuse it is cost hightly.Moreover,representative 1l minimization methods is discussed,and contrast two different representative 1l minimization methods in the experiments.(3)In chapter 4,firstly,the low-rank decomposition is introduced to align and denoise the facial images in the unconstrained environment.When the facial images of the unconstrained environment are classified directyly,the performance of facial gender recognition system will seriously decrease.Because there are a lot of disturbance and noise in the dataset of unconstrained environment.So the low-rank decomposition is introduced to align these facial images and denoise.In the phase of recognition,the collaborative representation mechanism,which substitutes1l-norm regularization with 2l-norm regularization to enhance the discriminant power,decrease the computational cost of the gender recognition system and solve the small-sample-size problem of facial gender recognition.At the end of this paper,LRDCR and other algorithms are contrasted in YouTube database which is a unconstrained set of facial images.LRDCR has more discriminative power than other methods.