An Effective Image Search Reranking Based On Prototype

Reranking is a common methodology used in all variety of fields and the same is applied here for images that are searched from web. The methods available for image search are based on text. The probability of containing irrelevant images in resulset , is more.The precision of text based image search result is improved by Image search reranking. Reranking is applied to rerank the retrieved images based on text surrounding the image, metadata and visual feature. Hence “Prototype Based reranking image search” is proposed here. The top ranked images are used as (noisy) training data and visual classifier is used to improve the ranking further. Given the keyword as input to the proposed system model, the output contains a set of reranked images which leads to increase in probability of exactness of user search requirement. The Principal novelty of overall proposed method is in combining text/metadata and visual features in order to achieve a completely automatic ranking of images. Primary assumption is that top images in the text-based search output are equivalently relevant and it is relaxed by linking the relevance of the images to their initial rank spot. Then to represent query visually and to construct Meta rerankers, employ a number of images from initial search result. From these results we can calculate reranking scores. These scores are then combined using a linear model to generate the final relevance score which is a new rank position for an image in reranking search results.