Return Entire Thread First 100 Posts Prev 100 Next 100 Last 50 Posts Report Thread Style: Yotsuba, Pseud0ch, Terminal

Pages: 1-

Tool for searching images


I've been looking for a tool that would search for picture's origins in the internet. If I have a picture on my computer and want to find out where is this picture from. What could I do, except asking ppl or trying to guess keywords in google? Is there a tool which could help me?


And how do you imagine this software should work?


I'll give full bag of internets to anyone who thinks it out.


If we search only unchanged pictures, then saving picture hashes and urls and searching by hashes would work perfectly. If we want to find slightly modified pictures as well (like saved .jpg->.gif or changed size), color histograms could be made of parts of image and then compared on similarity. While it's not character recognition system, it's not so difficult to do. Therefore I assume someone has already made the tool. Then, where is it? There HAS to be one!


>>4 is true, I (>>3) apologize. I am fucking idiot. But I still don't think you will find it. Serch engines don't manage (or just don't put out) file hashes and no-one else has capacity to crawl all those pages for a few users' profit.


It has been created for local usage (non-internet):

The technology was first written by a Swedish university student, and google will get it, just a matter of time.


Before google, I guess, the same imgSeek people will finish their own project of global search:
They have an interesting idea - to make the searching distributed. This is good because of many benefits:
- Search speed and db space distribution
- Many picture packages are shared only through p2p networks, not the www
- Web crawlers are slow/expensive

The bad side of distributed search is that doubtly many people will download and use such a tool, unless it will be included as a part of some bittorrent client.

But anyway, the project seems active and promising.


No that shit is abandoned and does not even work as intended. Stop trying to divert users to your stupid webpage. Please #7, please.


Damn! I was looking on "Last updated: Sat Jan 31 22:46:41 2004" and confused 2004 with 2005 :(
That leaves me to the one and only choice: full research of image searching as part of my studies. I will post here the results (good or bad).


I'm telling you right now. If you can manage to solve this problem and handle all possible cases, you will be a billionaire. I REALLY doubt imgseek people got anywhere with this.

The problem is not how the search is performed or even the time it takes. It is more about the accuracy of the results that are returned. There are just too many situations to account for outside of simple file hashes being correct. From sub images, crops, adjustments etc. even the slightest change ruins the hash values so now you have to get a machine to percieive the image as humans do.

If you want to look into this, study computer vision and you'll be exposed to alot more.


Sure, there is no way to handle all possible cases. Transformations like rotation are pretty impossible to handle in real time nowadays. But don't even hope to do that. What I am aiming at is unchanged images and images with slight transformations. And with slight transformations I will mainly aim at image resizing and saving in another file format. I don't hope much for transformations like sub images and crops. Still I will give even them a try because it's fun for me.

Most of the people look at the problem from a "do it all" perspective. And of course they find it impossible to do. I am looking at the problem from a "do it fast what you can" perspective. As I don't know any precedents, I don't know whether I will stop at being able to find only unchanged images or more can be done really fast. Either case, the result will be useful at least for me. And as I said: it's fun.

I have also considered the imgSeek's similarity to what I'm willing to make. That project is aimed from the very begining from picture finding from a rough sketch. Contrarary I will aim at pictures very similar to the target. Therefore even if their project went down this is not a downer for me. Because I will take a different path.

Well, you can still be sceptical about this, but I don't think I have much to add. There are two kind of people who hear about my problem: "go for it!" and "give it up!". Choose your side! :)


There is a major trade off between performance and accuracy when dealing with this particular problem. I know that the "do it all" perspective is unrealistic, but the problem is actually that difficult to implement an efficient, accurate solution using software ONLY. Realistically, it will require some form of human intervention along the way to produce very accurate results.

Finding identical images is the easiest part to work on but is is much more interesting to find sub images and resized ones too. If you want somewhere simple to start, try using the histogram of the image for comparison. This is fast and produces inaccurate results but it's a good starting point to see how these things work. And then it gets alot more complex than that...

Yes, I think you should go for it. It is a great idea but I don't think you know what you're getting into. Once you do more research I hope that you still feel as ambitious as you are now so good luck!

Newer Posts

Name: Email:
Entire Thread Last 50 Posts First 100 Posts Thread List Report Thread