Research Of Applying Dimensionality Reduction Algorithms And Relevance Feedback To Multimedia Image Retrieval

In recent years, with the development and popularization of multimedia and Internet technology, the large of multimedia data what user exposure and processing is increasing exponentially, such as image, video etc. Facing the growing large of multimedia data, how to achieve fast retrieval and improve retrieval accuracy has become a research and difficult subject. Multimedia image retrieval has been a hot and difficult research topic and received increasing attention in the field of machine learning and Computer Applications. Recently, There is a efficiency way that choosing an dimensionality reduction algorithm to improve the retrieval rate and using an appropriate relevance feedback technologyThis paper focuses on image retrieval based on PCA-LPP algorithm and relevance feedback method based on LPP. The goal of this paper is to improve the image retrieval performance of dimensionality reduction algorithms. The main contributions of this paper are as follows:1) This paper briefly introduces the Color, Texture future, Shape and Contour based on image Content and analysis the Classic dimensionality reduction algorithm, user’s relevance feedback.2) This paper researches three dimensionality reduction algorithm, PCA(Principal Component Analysis),LE(Laplacian eigenmap),LPP(Locality Preserving Projections). Though compare the effect of the parameters and image retrieval experiment, analyze the advantages and disadvantages of the three algorithms, we propose a novel dimensionality reduction algorithm called PCA-LPP. Different form PCA which aims at preserving the global Euclidean structure, it aims at preserving the local neighborhood structure on image dataset. To compare with the LPP, it can remarkably improve the retrieval recall rate. Therefore, PCA-LPP is less sensitive to outliers than PCA and preserveing the local manifold structure. Also, comparing to three algorithms such as LE, PCA, LPP, PCA-LPP has better performance in image retrieval.3) Though research LPP and relevance feedback based on neural network, we propose a new algorithm called RFLPP. In order to improve the efficiency of retrieval accuracy, the article incorporates user’s feedbacks. Using the algorithm of LPP, we map the data points to a subspace. In this subspace, a weighted graph G can be constructed by a candidate data set which to consist of k nearest neighbors of query data points, and query data set. We then compute the geodesic distances between all pairs of vertices of the graph G, and sort them, obtain feedback results.Experimental results on famous image datasets (CorelIK and Croel5K) in Windows7by the Matlab software show that the proposed algorithms PCA-LPP and RFLPP substantially improve the retrieval performance for image retrieval.