Browse Prior Art Database

A Method and Apparatus of Automatically Assigning Images to Plain Text in Searching Results

IP.com Disclosure Number: IPCOM000193081D
Original Publication Date: 2010-Feb-10
Included in the Prior Art Database: 2010-Feb-10
Document File: 3 page(s) / 39K

Publishing Venue

IBM

Abstract

This article is for automatically assigning images to plain text in searching results. The main process consists of three steps: 1) Focused Entities Extraction which is a component of finding most topical named entities among all entities in a document; 2) Image Retrieval which is a component to search the relevant images in an image database; and 3) Similarity Metric which is a component to metric both the image similarity and the text similarity. After these steps, the most relevant images is assigned to the corresponding plain text.

This text was extracted from a PDF file.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 55% of the total text.

Page 1 of 3

A Method and Apparatus of Automatically Assigning Images to Plain Text in Searching Results

Nowadays, search engines are becoming more and moreimportant for us to extract valuable information. Some general search engines have already achieved remarkable success.

As we know, images are more intuitive than text in human reading. However,

no existing search engine has the feature of assigning images to plain text in searching results.

Imagine that you want to buy a cell phone and are referred to Alcatel-735i by a friend. You put the keyword 735i (if you don't know how to spell Alcatel or just forget it) to a searchengine to get some reviews. Unfortunately the search results contain at least three products: BMW 735i (car), HP 735i (recorder) and Alcatel 735i (cell phone). For this case assigning images to plain text in searching results can help you review the products much faster.

The aforementioned example is too simple to frame this issue. Let's take Toyota Center for another example. You want to know something about Toyota Center, the home field of Houston Rockets. However, the searching results may consist of two topics: the field and the match. You have to read the snippets of searching results to distinguish which are useful. If each searching result has been assigned a image about the field or the key figure of the match, you can easily get the information you want.

Though assigning images to plain text in searching results is commercially valuable, it is not an easytask. How to construct the semantic relationship between plain text and image?

So the key problems are clear:
a) How to improve the sear...