Monday, July 14, 2008

A few thoughts on image retrieval

"I’ve been accused of finding the photo of Andrew Moro (In the opinion of a few…really NOT Andrew Moro I suppose) and creating his Facebook profile in order to promote my “smear” campaign.
They can see that the guy in the photo looks remarkably like Liz Moro. The resemblance is uncanny.
The question they need to answer is…
How did I find it?How did I find a photo of a guy who looks like Liz Moro?
Undoubtedly, they all have a basic understanding of how image retrieval works. This is from Google:
How does image search work?Google analyzes the text on the page adjacent to the image, the image caption and dozens of other factors to determine the image content. Google also uses sophisticated algorithms to remove duplicates and ensure that the highest quality images are presented first in your results.
Google uses a text-based search to retrieve images. Important deciding factors in determining whether or not an image is relevant for a particular search include the name of the image and the surrounding text.
In other words, I must have used the photo’s surrounding text to retrieve it.
Because they assume that this photo isn’t Andrew Moro I couldn’t have used Google’s image search to find photos of “Andrew Moro”.
However, this guy looks exactly like someone named Liz Moro and presumably that’s what I wanted.
Given that, what are some possible search techniques I might have used to retrieve this particular photo?
I might have tried to find a photo of Liz Moro online and hope that there would be family members close-by.
They know there aren’t any photos like that online.
I might have assumed that “Moro” was an Italian surname and then used the image search feature to search for “Italian men”.
I invite someone to try this and see if they find the photo of Andrew Moro.
Another technique would be to search for random photos with .jpg extensions. One could then sift through hundreds of millions of photos searching for a young man who strongly resembles Liz Moro.
I just tried that and I retrieved 562,000,000 results.
The monkeys will type Shakespeare before I go through that and find what I want.
Of course, if I found it…it’s out there for anyone to find, right?
Remember, you can’t use “Andrew Moro” or “Moro” because I supposedly created the Facebook account with someone else’s photo.
So, get to work!"

Saturday, July 12, 2008

Image Retrieval on Facebook !!

img(Finder) is an experimental CBIR(Content Based Image Retrieval) desktop application, that retrieves images from Facebook.The user is connected to Facebook using his personal account data and the application downloads information from the images of the user and his/her friends albums. The data are used to index the total of the images.

Current Version: - Ver 0.1 beta
- Connects to Facebook.
- Indexing the albums of the user's friends.
- Image Retrieval WITH SKETCHES.
- Image retrieval from the albums based in visual example.
- Browse the images in your friend's albums
- -NEW- .Net Framework 2 compatible
- Now with the newer version of CEDD AND FCTH


To Do:
-Improve the application's speed.(The current version is too slow)
-Keyword based image retrieval.

Download img(Finder) - Ver 0.15 beta -- Img(Finder) is now embedded in img(Rummager)
Windows Application
.NET Framework 2

For any suggestions, comments or bugs please contact me directly.
NEW!! The application is open source. Contact me in order to send you the source code.


Wednesday, July 9, 2008


Please Update your bookmarks: New url

Tuesday, July 8, 2008

LIRe: Lucene Image Retrieval

Mathias Lux and Savvas Chatzichristofis paper "LIRe: Lucene Image Retrieval" has been accepted to the ACM Multimedia 2008 Open Source Software Competition.

LIRe is a Java library for content based image retrieval. LIRe extracts image features from raster images and stores them in a Lucene index for later retrieval. LIRe also provides means for searching the index. LIRe is intended for developers, who want to integrate content based image retrieval features in their applications. Due to the simplicity of the approach (no database and only a few lines of code are needed to integrate LIRe) it is an easy way to test the capabilities of classical CBIR approaches for single application domains. Also the integration of additional image features is possible easily to further extend the functionality of LIRe. Currently the following image features are included in LIRe:

Basic color histograms in RGB and HSV
MPEG-7 descriptors scalable color, color layout and edge histogram.
Tamura Features
Color and edge directivity descriptor, CEDD
Huang’s AutoColorCorrelation feature