Millions Advertisings Homepage.png)
10
A few words mean more than an image
Filed Under (Information Technology) by admin on 10-10-2008
The saying states that a picture can assert more than a thousand words. But that only applies to us humans.
When it comes to computers and search engines, a single word can make the difference between finding the right image and let the wealth of information concentrated in a picture or drawing lost in the midst of a multitude of other images less relevant.
Extracting textual information of images
Dozens of attempts have been made to extract information from textual information contained in visual images. The most efficient, however, still rely on the texts inserted by humans themselves as title or description as those images.
Now, researchers believe Europeans have an important step in the automation of that task, as well as towards the ability to retrieve information from images for which titles and descriptions were not properly allocated.
Layer information
The solution is the creation of a layer of information superimposed on images from digital files. That extra layer informational text will utilize the data entered by users, but will also be able to generate additional data automatically.
Because this is an extremely complex area, the researchers of the project aceMedia reused several techniques already developed. This applies to the identification of keywords visual low level - as areas of consistent colors that can be interpreted as sky, sea, sand or snow.
The same technique allows the identification of shape, texture and shape.
Key words and fields ontological
Combining the keywords with sets of rules contextual kept in areas ontological - such as the fact that sand and snow usually do not appear on the same image or that blue areas on top of an image are very likely to be the sky - the researchers managed transforming the digital data into images of a rich source of information.
Data for keywords are also low-level combined with results from specific detectors, such as programs able to locate faces in a photo.
Personal Rules
Users can also add your own preferences to the layer of information, defining requirements for allocation of keywords and create a system of personal archive of images that allow their subsequent recovery.
The platform developed by the researchers was named ACE (Autonomous Content Entity). The ACE layer defines an API that supports connections on a network, management of databases, coding, scalable, pre-processing of content, viewing the content, content analysis based on knowledge, as well as modules for modeling and contextual analysis.
Full Feed RSS
