Profile cover photo
Profile photo
Pic Lookup
About
Posts

Post has attachment

The R&D department at PicLookup has been misbehaving, as usual. My focus is supposed to be on our search site, trying to generate traffic, and on getting feedback from anyone and everyone on what they think our image search technology could be used for. But instead, I'm working to get our mascot, Pic, able to move about using his stereo cameras.

As you may infer from his picture, the two cameras gather stereo images, from which the distance of objects can be calculated. I won't go into a tech talk drawl here, lest it get too boring. But nowadays everybody seems to be using OpenCV for calibrating cameras and calculating range (distance to things in the world).

Why do it the standard way, when it's more interesting to strike out on your own, and try something simple that could actually work? Instead of all those lens distortion coefficients and pictures of chessboards, and I'm talking about the tried and true technique here, I'm trying to do the same thing, basically, but using a little bit of image search technology instead.

I'm tracking objects in the view as the camera servos move the focus around. And by knowing how far the cameras move, and how far the tracked object appears to move, I can (with a bit of difficulty) arrive at an expression for pan and tilt distortion (by the camera) for each region of the image. Why? Well, you need that to help calculate distance of things in the image. An object's range, or distance is directly proportional to the difference of its location in the left versus right eye (er, camera, that is). However, the camera lens distorts things a bit, so you want to correct for that, first.

So far though, it looks kind of unnecessary, since the distortion figures I'm getting are very small. But it's been a fun challenge so far, like everything else about building a robot.

And back to my appointed task: if you've had a look at our PicLookup site, feel extremely welcome to offer any feedback, whatsoever. That way, it'll look like I'm doing the job I'm supposed to be doing.

Or, drop me a line about anything else, like robotics, or AI, or why is everybody using mobile phones these days, anyway? My wife just got a Droid, and I haven't figured much out yet.
Add a comment...

Post has attachment

Post has attachment
If you're interested in image processing and image search, and you want to quickly get up to speed with current technology, take a look at PyImageSearch, and tell Adrian we sent you! He'll have you doing stuff like face detection and OCR in no time. Seriously, if you like to learn by doing, he's got cookbooks to get you started and offers advanced training from there. We're talking 15 lines of code, and you're detecting objects! And if you like theory, you get to learn about the state of the art, and can then deepen your knowledge in areas important to you.

At PicLookup, we're pretty good at fast search, but we're getting interested in AI, Deep Learning, and all that stuff, which is how we found Adrian's site.
Add a comment...

Post has attachment
Add a comment...

Post has attachment
This is from another great site, and covers reverse image search. PicLookup is perfect for this kind of forensic search, since the original image can be found from a mere patch taken from it.
Add a comment...

Post has attachment
Fantastic article, clear, informative and relevant. This is the kind of thing PicLookup was made for, which can find the source using only a tiny patch, or cutout of the original image. See piclookup.com/faces.html, which shows how to foil a catfisher's fake selfie photo.
Add a comment...

Post has attachment
Add a comment...

Post has attachment
Imagine that someone posted an image of people at a party, and somebody else copied a piece of that picture, and posted it on their own site, saying it was himself/herself. How would you ever know it was fake? You can't search for the original using just that stolen portion. Not with today's image search engines. Why not? Because the big search engines usually assume that you're uploading the entire picture for search, so they wouldn't find any images matching that tiny stolen "portion" of the picture that got posted.

But picLookup can find it! How so? We scan images in great detail, and we store it all. That allows us to find your original image from a small clipping, since we'll recognize the clipping itself, and then know where it came from (the original party image).

Why don't the big guys do this too? Well, gathering all that information from an image and storing it is expensive. And they'd also have to rewrite their search program to do this. Why bother? Nobody today seems to expect or demand this ability. We're just glad to be able to search for images. Right?

Probably so, but it will change, and the sooner the better, we say here at picLookup. Computers are faster and cheaper every minute. Our technology is here now, and it gets cheaper all the time.

There's more, too. Today there is a huge focus on computers understanding what the picture is of. Someday, perhaps soon, the image search engines will have an idea of what they're looking at (your uploaded image), and that could help them find a match.

But if you're using a small portion of an original image, a snippet, and trying to find the exact original, we're already good at that right now. And it doesn't have to be just portraits. Our search engine is quite dumb! You could upload a snippet containing a section of bark on a tree in a picture, and we'd match it, even if the snippet has been rotated and re-sized.

I'm not sure an image-understanding search engine would help with that. Bottom line, we have this now, and we just need folks to look it over and give us some comments, like 'great' or 'useless' (not!), or 'have you thought of this...'. Then we'll pull out the stops and rapidly scale up.

If you've read this far, thanks for your interest. If you're feeling lazy, we have a two minute video on youtube that shows some searches. (See https://youtu.be/7aAWEdGh70g)
Add a comment...
Wait while more posts are being loaded