ResearchBib Share Your Research, Maximize Your Social Impacts
Sign for Notice Everyday Sign up >> Login

Efficient Retrieval of Images for Search Engine by Visual Similarity and Re Ranking

Journal: International Journal of Advanced Computer Research (IJACR) (Vol.3, No. 10)

Publication Date:

Authors : ;

Page : 47-52

Keywords : Image search; Image Retrieval; Efficient Image Search; Image Re ranking;

Source : Downloadexternal Find it from : Google Scholarexternal

Abstract

Nowadays, web scale image search engines (e.g. Google Image Search, Microsoft Live Image Search) rely almost purely on surrounding text features. U sers type keywords in hope of finding a certain type of images. The search engine returns thousands of images ranked by the text keywords extracted from the surrounding text. However, many of returned images are noisy, disorganized, or irrelevant. Even Goo gle and Microsoft have no Visual Information for searching of images. Using visual information to re rank and improve text based image search results is the idea. This improves the precision of the text based image search ranking by incorporating the infor mation c onveyed by the visual modality. The typical assumption that the top - images in the text - based search result are equally relevant is relaxed by linking the relevance of the images to their initial rank positions. Then, a number of images from the in itial search result are employed as the prototypes that serve to visually represent the query and that are subsequently used to construct meta re rankers .i.e. The most relevant images are found by visual similarity and the average scores are calculated. B y applying different meta re rankers to an image from the initial result, re ranking scores are generated, which are then used to find the new rank position for an image in the re ranked search result. Human supervision is introduced to learn the model wei ghts offline, prior to the online re ranking process. While model learning requires manual labelling of the results for a few queries, the resulting model is query independent and therefore applicable to any other query. The experimental results on a repre sentative web image search dataset comprising 353 queries demonstrate that the proposed method outperforms the existing supervised and unsupervised Re ranking approaches. Moreover, it improves the performance over the text - based image search engine by more than 25.48%

Last modified: 2014-11-28 22:04:38