What is even more impressive is that JSFeat doesn't just implement the simple image processing you find in other libraries - it also does some cutting-edge object tracking and detection.
The basic operations that have demos are:
- Basic image processing methods (grayscale, derivatives, box-blur, resample, etc.)
- box blur
- gaussian blur
- equalize histogram
The more advanced operations are:
- canny edges
- Fast Corners feature detector
- Lucas-Kanade optical flow
- HAAR object detector
- BBF object detector
The all of these work in realtime from a web cam.
The Lucas-Kanade optical flow method is a very basic form of object tracker and in this demonstration it works well. All you have to do is click on a point that defines an area you would like to track and the point stays put as the image moves. Occasionally the image movement is too great and the point will lose the track. This is the basis of more advanced object trackers such as Predetor. All of the others are worth trying out, but the two object detectors customized as face detectors are particularly impressive.
Canny Edge Detection
Another really good thing about JSFeat is that its documentation is very good - way better than anything you would normally find in an open source project. It is all the work of one programmer, Eugene Zatepyakin, aka inspirit.
What does this mean for the future of what can be done in the browser?
As long as the other browsers catch up with Chrome and implement WebRTC fully, we should be able to have some interesting AI image processing applications live and without the need to offload the processing to a server. The JSFeat implementations would make a good starting point for a lot of interesting and adventurous web apps.