Font Size: a A A

Manifold Learning And Dictionary Learning Based Image Retrieval

Posted on:2015-09-23Degree:MasterType:Thesis
Country:ChinaCandidate:L J YanFull Text:PDF
GTID:2308330464968545Subject:Communication and Information System
Abstract/Summary:PDF Full Text Request
In recent years, with the rapid development of Internet and the continuous improvement of computer processing performance, image data is growing at the rate of swelling every day. Now the question is how to quickly and accurately find the images from the vast amount of image data that meet user’s requirements, to solve this question an efficient image retrieval technique is needed. Traditional text-based image retrieval technique has the disadvantages that the amount of manual annotation labor is large and manual annotation has a certain degree of subjectivity. After the 1990 s, Content-Based Image Retrieval(CBIR) technique has become a hot research field, currently image retrieval with relevance feedback technique becomes the focus of research. Starting from manifold learning and dictionary learning, we give an in-depth study of image retrieval based on relevance feedback technique. The main work is as follows:First, despite the success, using Maximum Margin Projection(MMP) algorithm in Content-Based Image Retrieval(CBIR) still has drawbacks. It treats the positive and negative feedback samples equally, which is not appropriate since the two groups of feedback samples have distinct properties. To overcome the deficiency of MMP, we propose the Biased Maximum Margin Projection(BMMP) algorithm. The algorithm treats the positive and negative feedback samples asymmetrically, and it can better explore the geometric structure information of the data by keeping the local structure of unlabeled samples in the neighborhood of feedback samples. The experimental results show the performance of BMMP algorithm.Second, the research shows that dictionary obtained from the training samples by machine learning methods has the advantages of good flexibility and high adaptability. K-SVD is a widely used dictionary learning method, since the training samples can have a good representation in the leaned dictionary. This method has achieved good performance in the areas of classification and recognition. Based on the above analysis, we put forward K-SVD dictionary learning based image retrieval method. This method can better reflect the user’s query intention by iteratively updating the dictionary with the increase of feedback information. The experimental results show that this method is superior in Content-Based Image Retrieval.
Keywords/Search Tags:Image Retrieval, Relevance Feedback, Local Geometric Structure, Dictionary Learning
PDF Full Text Request
Related items