Does anyone know what app/research this is? I can't imagine why smartphone maker don't use this for their crappy camera's so they can still embed cheap camera's but enhance the picture quality.
[0] https://docs.opencv.org/2.4/doc/tutorials/calib3d/camera_cal...
The area of research that covers from HDR to smartphone array cameras is called Computational Photography, I think.
It shouldn't take too much effort to implement the correction data in the camera software.
On the one hand, Lightroom corrects my lens distortion and vignette from a saved profile of that lens; but I’d think at a smartphone level, where the parts are smaller, they’re also less even?
In other words, camera lenses are small so manufacturing defects are more noticeable. Therefore, each smartphone requires a different correction.
I don’t know, I’m just guessing.
> In other words, camera lenses are small so manufacturing defects are more noticeable. Therefore, each smartphone requires a different correction.
Well, yes, and no, and yes. Small camera lenses are probably going to be more evident in certain defects. On the other hand, there are typically less elements in smaller camera lenses, which is nice as the less of those there are, the less the chance of other defects.
Taking a picture like mentioned above could absolutely create a 'per-camera/phone' correction profile. May not be able to correct every type of defect but I could see it being useful for some lenses. I know for Sony cameras, there have been a couple models where a large percentage of the copies produced have -one- soft corner.
* Vignetting always improves
* First-order distortion improves, second-order distortion sometimes gets worse.
* Corner sharpness usually improves, unless midway out on the frame was the weakest performance on full frame.
* Center sharpness (image-scale) drops due to the additional enlargement.
* Lateral chromatic aberration generally doesn't get significantly better or worse.
* Longitudinal chromatic aberration becomes more visible because of the additional enlargement.
Nitpick, that's just camera calibration. No ML needed.
This usually provides acceptable results for applications like correcting lens distortion, even across multiple instances of the same lens. However, sometimes better accuracy is needed - either because the lens is very cheap and inconsistent between different examples or for applications like SLAM where better camera calibration translates directly into better results. In that case there are techniques like online calibration to tweak the calibration parameters on the fly.