New Optimization Method For Collaborative Representation And Its Applications In Image Recognition

Image classification based on sparse representation is a kind of new way which uses compressive sensing to it and it has higher classification performance.Collaborative representation algorithm is an important branch of sparse representation and its classification property is often affected by the redundancy of sample data.Recently,some latest papers use more effective training samples or their characteristic to construct completed dictionary for representing the test sample with the purpose of improving the classification performance.Inspired form it,on the basis of mining more effective face features,aiming at optimization learning of dictionary,we propose optimization representation learning method which uses facial features as elements of dictionary for better classification result.The main work and innovation points are summarized as follows.1)On the optimization learning of dictionary: we propose collaborative representationbased classification based on optimization of dictionary.The key motivation of proposed method is to exploit more similar training samples of test sample through histogram measurement in order to reduce uncertainty from the initial data set.We first obtain histogram vectors of training and test sample,then measure similarity of test and training samples in the low dimension space of histogram by Euclidean distance for the sake of getting more competitive training samples.Finally these achieved training samples are used as elements of the dictionary to construct a better representation of test sample for robust face recognition.Experimental results conducted on three commonly used face databases including ORL,FERET and Georgia Tech demonstrate its feasibility and effectiveness.2)On the structure of feature dictionary and optimization learning: we put forward collaborative representation-based classification based on structure of LBP dictionary and optimization learning.It uses local features of samples instead of samples to construct a completed dictionary for enhancing robustness against facial changes.The method gets local binary pattern features in blocks and uses collaborative representation method for classification.We conduct extensive experiments on publicly available database to verify of the proposed algorithm.Meanwhile,In order to reduce the redundant of feature dictionary,we apply dictionary optimization method into feature dictionary.In this way,we get advantage of nice classification performance of LBP and reduce the redundancy of features dictionary at the same time.It is conducted on two commonly used face database and the result shows that there is a further increase in recognition rate.But our method consumes more time because of feature extraction and elimination strategy,i.e.it raises recognition rate at expense of more time.From what has been discussed above,we do some research on the basis of CRC and put forward some improvement schemes.To some extent the recognition rate is raised.