[BBC-Micro] Reversing the Tube ULA (destructively)
percy.p.person at gmail.com
Sat Oct 30 16:47:32 BST 2010
Christian is OK for me to forward these tips to the list:
On 30 October 2010 16:26, Christian Sattler <sattler.christian at gmail.com> wrote:
> My tools are based on autopano SIFT
> (http://en.wikipedia.org/wiki/Scale-invariant_feature_transform) for finding
> many control points between the images (around 10^5 for 100 images). A
> substantial part of those will be very wrong, so I filter them using
> statistical methods, which also estimate image positions and optical
> distortion parameters along the way. Some notes you should consider before
> 1) I can only stitch images which are non-rotated with respect to each other
> (not a problem for most microscopes).
> 2) Depending on the amount of detail in the images, I usually need 100-200
> pixels of overlap (more is better, of course). With your 6MP camera, I'd say
> go for at least 200 pixels.
> 3) If you can, get the RAW image sensor data, where partially destructive
> filtering, processing, and interpolation (usually, 2/3 of the pixel data is
> pure guesswork of the camera) has not yet been applied (you can always do
> the processing afterwards). This will also enable correction for transversal
> chromatic aberration if present. Also note that usually the RAW will include
> more pixels at the borders, increasing overlaps.
> 4) Probably not that significant, since 20x will suit all your needs: A too
> short exposure time will introduce noise.
> 5) Try to stay in the same focal plane with respect to the chip surface
> (level it before shooting).
More information about the bbc-micro