Research Of Image Retrieval Algorithm And Its Application On The Internet Education

SIFT is commonly used feature extraction method in image retrieval field, butthe threshold and the dimension of feature vector have a greater impact on theretrieval accuracy and efficiency; on the other hand, for the practical application oflarge scale image retrieval on Internet education online, we need to consider retrievalaccuracy and efficiency so that the user has a good interactive experience.This paper studies these problems. Considering textbook’s image has mainlyword, little change in color, different image brightness and probably partial image,SIFT feature descriptor has rotation, scaling and brightness in-variance, we adopt theexcellent performance characteristics of the local shape feature-SIFT when extractingimage feature. The article’s main research work are as follows:(1) For the problem of partial testing images hasn’t returned the ideal image byextracting SIFT feature points, we reduced the threshold and got more features tomatch. This paper introduced the change of threshold has an impact to the number offeature point and the matching points between two images. We proposed the viewabout the number of feature point determine the retrieval accuracy and how to applyin image retrieval field.(2) There will be an issue of high computational complexity, long retrieval timewhen do exhaustive search in thousands of pictures,this paper proposes a fast retrievalmethod based on clustering. The algorithm executes fisher coding using the bag ofvisual words, i.e. Fisher vectorization image. Firstly, this paper uses an unsupervisedclusting method to mark fisher vectorization images on each category. Secondly, Wefind the nearest category to match this image’s features and the features of the nearestcategory which this image belongs to, and get TOP3 images finally. Experimentsshow that the proposed method can narrow your searching field, reduce the retrievaltime and also ensure the retrieval accuracy.(3) Based on these studies, We introduce the improved algorithm how to applyimage retrieval module based on big data technology, for example, how to train imagedatabase and adjust the number of feature point to image retrieval.