Weighted Covariance Representation For Image Set Classification

It is popular to share photos and videos with smart phones or other digital devices, which indicates the convenience in obtaining visual data. A great quantity of available visualized database prompts us to deal data with sets. Traditional image classification problem, as a hot issue in computer vision and pattern recognition, has developed into set-based classification problem in recent years. The keys of solving the problem are image representation and similarity metric between image sets.In this paper, we propose a novel weighted covariance representation for image set. The representation not only describes the image set with column structure, but also deals with noisy data because of the singular values. Secondly, we obtain the column basis via dimension reduction method. At the same time of obtaining the basis, we have already get rid of the influence of unfavorable factors such as noise. Also, we discuss the similarity metric in this paper.Experimental results show that the proposed weighted covariance representation succeed in recognize the labels in image set classification.