For Project Enquiry +91 9962 067 067

EXISTING SYSTEM

PROPOSED SYSTEM

EXISTING CONCEPT:-

In existing process implemented by indexing and searching the textual information associated with images, such as image file names, surrounding texts and universal resource locator.

Text-based search is first applied to obtain a coarse result from a large text-indexed image database. Then the top returned images are reordered via various re-ranking approaches by mining their visual patterns.

PROPOSED CONCEPT: -

We propose a new diversified re-ranking method, TARerank, to refine text-based image search results. This method directly learns a re-ranking model by optimizing criterion related to re-ranking performance in terms of both relevance and diversity in one stage simultaneously.

The query dependent features for each image are extracted from textual information to describe the relationship between a query and an image.

EXISTING TECHNIQUE:-

Relevance-based and diversified re-ranking methods.

PROPOSED TECHNIQUE:-

Topic-aware re-ranking (TARerank).

TECHNIQUE DEFINITION:-

Relevance-based re-ranking is to maximize the relevance of the returned image list through reordering. Maximizing the relevance of each item in the list is the only objective.

The resulting ranking list tends to return a large number of redundant images that convey repetitive information.

TECHNIQUE DEFINITION:-

Topic aware re-ranking is proposed as learning based re-ranking method. It directly learns a model from a training set by jointly optimizing relevance and diversity.

Introducing a learning procedure, the gap between low-level visual feature diversity and high-level semantic topic diversity is bridged to some extent effectiveness of our method.

DRAWBACKS:-

Relevance-based re-ranking and diversified re-ranking do not capture the hierarchical topic structure of search results very well.

In this model cannot integrate visual features, which are efficient in refining the click-based search results.

ADVANTAGES:-

Visual and click information are simultaneously utilized in the learning process for ranking.

Accurate ranking model can be learned from this framework because the noises in click features will be removed by the visual content.