I do not know what autopano really does but I would try to convert internally to grayscale images, down-rasterize their resolution so that only differences in light/color intensities would show up. This could be used to delineate regions which are distinct from their surroundings. A pattern of such raster-pixels could be used to quickly find regions of overlap.

I already sent an email to Alexandre J about this but let me repeat that I vote for introduction of a measure to calculate how many CPs are in a region. To my taste too many CPs are located in clusters and I wonder whether they are useful at all (and other in forum already complained on this as well). Especially if you think most of them in the cloud are "green" while some in the middle are "brown". Optimizer does not optimize position of these "brown" CPs at all although clearly those brown-ones surrounded by green-ones were misplaced. I wish it is relatively easy to fine-tune such misplaced CPs. Then, CPs closer than some number of pixels could be dropped. Preferably, autopano should pre-filter CPs and find those in areas where is low error likelihood for misplacement. For example, position in the middle of a thick spot should be preferred over a CP in a narrow spot, which might be smaller/lost on another exposure, or its boundary altered. CPs in regions locally uniform (evenly colored surfaces, e.g. car, boat, green apple) should be preferred over globally uniform regions (like the ocean, sky, wooden table). Finally, autopano could look for evenly colored objects in the pictures and try to use them as patterns of CPs. For example imagine a ladder in a picture with some random noise behind it (grass, trees, soil, whatever "small" objects).

Certainly, I support the idea of somebody who posted into the forum topic "Future" showing problems wit assembly of two images with Rambo, some green meadow in the backgroung and some electricity cables hanging in the sky (which showed nicely where the CPs have to be present). That is similar to my issue with roof of a house not having a straight line.

Many times in the forum somebody opted for preference for CP in corners and I support it. In my experience I find better stitching with DISABLED lens corrections (Olympus C5050z) than with enabled correction. For example, edge of a roof of of house is broken at the junction when lens correction is enabled. That feature decreases the RMS value but other than that does nothing good for me.

Last edited by mmokrejs (2009-12-30 22:03:31)