Comment 1
Comment 2
References:
Title: Cosaliency: Where People Look When Comparing Images
Authors: David E. Jacobs, Dan B. Goldman, and Eli Shechtman
Venue: UIST 2010, Oct 3-6, 2010
Summary:In this paper, the authors discuss a computerized method of locating the changes between a pair of images. This algorithm would be useful in a situation they call "image triage," in which a photographer needs to decide which images to keep and which images to delete on a camera, usually to open space for more photos. To assist with the design of this algorithm, they come up with a concept called cosaliency, which is similar to image saliency, where you are trying to find the important part of an image, but instead between two pictures.
They began by creating a research project on Amazon Mechanical Turk wherein they asked the workers to compare series of image pairs and select small crops of the images that they thought had the biggest changes in them. From this data, they created an equation to model this behavior, and then applied it in their software for cropping. Finally, they generated images cropped by this algorithm and again used Amazon Mechanical Turk to determine how much people liked them. In general, they found that the users liked it more than traditional methods.
Discussion:
I liked this article because it used hard numerical data to back up its points. Even though it used a lot of jargon that I was unfamiliar with, the equations they showed were all solid and it provided plenty of images to demonstrate the process. For this reason, I am confident in the usefulness of this paper.
From a content perspective, I am not a huge photographer, but I can see how useful this would be to people who do like to take photos. This and many of the other works that they mentioned as they created their new algorithm will surely help make the newer model digital cameras even easier to use.
I have to agree, this was a very technical paper, but overall, the solution to the problem seems promising. I would like to see more research done on more than just pairs and focus more on dynamic content. They might have to look at frame skipping algorithms used in video transcoding.
ReplyDeleteI also liked all of the data that backed up the paper from the algorithms to the studies that were done with users. I think even if this idea doesn't work out to its full potential the algorithms developed could possibly be helpful in other applications.
ReplyDeleteI found it very interesting that they used Amazon's Mechanical Turk service to gather their base data. While I'm not sure how vital it is to develop technology to help save memory card space on a camera, with storage technology always constantly getting better, this was a very fascinating paper none the less.
ReplyDelete