This interesting article explores pros, cons, and challenges of face recognition - how it can be used for good to find criminals quickly, and for bad, potentially allowing google glasses to identify a stranger and give you his social security number.
And here's a blog that links to some examples of super-resolution video processing to create a high resolution image from a grainy one.
- An example with nameplates, showing extreme reconstructions
- Hallucinating faces
- An opensource implementation of video-based SR and a commercial multiple-image SR implementation
- Video to video SR, and another one...
- Classic paper of single-image super-resolution, trying to find similarities in the same image (to a degree, similar to fractal methods for image compression, which can indeed be used also for SR). Less strong results as the algorithm makes no prior assumptions.
- Multiple cameras (i.e. surveilance)
- More and more...
And here's an interesting proposal that you can increase a camera's resolution by shaking it. In essence, an indistinct line that crosses two pixels on the recording chip is made more distinct by shaking the chip by an amount less than a pixel width and then analyzing the consecutive images to achieve better clarity.
Sent from my iPhone
No comments:
Post a Comment