Cover image
project image

Web Image Re-Ranking by utilizing Text and visual features

Archana Singh




Price : $ 2

View : 188      Download : 8

Product Discription


Abstract— Images are the increase day by day on the Internet, retrieving relevant images from a large collection of database images has become an important research topic. This paper focus on the reranking of images by utilizing the both the visual and textual features. So given a textual query in traditional image retrieval, relevant images are to be re-ranked using visual features after the initial text-based search. Here first query keywords are utilize for separating the dataset images into two group of relevant image and irrelevant image then all the images are ranked base on the image different modality of image features as the similar images need to be display closer. Using single modality is not effective as different image need different kind of feature for analysis and it was obtained in experimental that the proposed re-ranking approach has better performanceis than using single modality.


There are 7 global features extracted, including

  • 225-dimensional Block-wise color moments. Each image is split into 5-by-5 blocks, and 9-dimensional color moment features are extracted from each block.
  • 64-dimensional HSV color histogram. A 64-dimensional histogram feature vector is extracted in HSV color space for each image.
  • 144-dimensional Color autocorrelogram. HSV color moments are quantized into 36 bins with 4 different pixels pair distances.
  • 256-dimensional RGB color histogram. A 256-dimensional histogram feature vector is extracted in RGB color space.
  • 75-dimensional Edge distribution histogram. Each image is divided into 5 blocks and 15-dimensional EDH features are extracted.
  •  128-dimensional Wavelet texture. 128-dimensional features are extracted using the mean and standard deviation of the energy distribution of each sub-band at different levels.
  •  7-dimensional Face features. The features include the number of faces, the ratio of face areas and the position of the largest face region.

Conclusions: World Wide Web has necessitated the users to make use of automated tools to locate desired information resources and to follow. Web image re-ranking has been widely used to reduce the user searching time on the internet; its success mainly depend on the accuracy of image features similarities. This paper present utilizing of the new text as well as visual features for ranking the image as both make the re-ranking process more powerful, which is shown in results. 

Project Sample Image

Other Detail

Software Requirement :   MATLAB 2012 and Above

Hardware Requirement :   RAM: 2GB, Processor 2.2Ghz

Application :  

Project Attachement

PDF IEEE Base Paper
Doc Dissertation Complete Document
Source Code Complete Code files