[ptx] Enblend performance with varying exposure values

[alexandre jenny]

I've already planned to create a HDR enblend, that is a rendering
system which will use many pictures with different exposures to create
a panorama directly in HDR format by using shutter speed and exposure value.
It should not be really complex to create by following what Paul debevec did
for it hdrshop.

Alexandre Jenny


> -----Message d'origine-----
> De : ptx-bounces at email-lists.org 
> [mailto:ptx-bounces at email-lists.org] De la part de JD Smith
> Envoyé : mardi 27 avril 2004 19:19
> À : ptx at email-lists.org; PanoTools at yahoogroups.com
> Objet : [ptx] Enblend performance with varying exposure values
> 
> 
> 
> For those users of enblend out there: have we reached the 
> point that exposure lock is no longer necessary?  That is, 
> can you fix the focus, and possibly the aperture (for good 
> depth of field and/or resolution), but otherwise let the 
> camera meter its own desired shutter speed?  I'd be 
> interested in seeing example panoramas created with varying 
> exposure values from image to image, and blended with 
> enblend.  This obviously has the capacity to yield panoramas 
> with higher dynamic range (think of including the sun along 
> with deep shadows opposite the sun).  
> 
> I've also been interested in a hypothetical camera "panorama 
> mode" which doesn't lock the exposure value (read shutter 
> speed) as is traditional, but does remember the last value 
> from the previous exposure.  It will allow the EV to change, 
> but will not let it change by more than, say, 1/3EV (about 
> 25% more or less light admitted) from frame to frame.  This 
> would help the final blend look smooth, but would also extend 
> the dynamic range smoothly from bright to faint portions of 
> the panorama.  
> 
> The result would obviously depend on the order in which you took the
> panorama: starting out with the sun and moving off into the 
> shadows, it would take several frames to bring the EV up 
> enough to compensate, but the change will be smooth and 
> gradual, especially for panoramas formed with lots of images 
> (e.g. 30 degree FOV lens).
> 
> Another possibility would be to divide the image data by the 
> exposure time (assuming the aperture is fixed) from the 
> header, before stitching and blending.  While this won't help 
> with clipped highlights, it should normalize the 
> color/intensity of smooth regions like the sky.  I suspect, 
> however, that the place this scaling should take place is in 
> the RAW data, before the Bayer interpolation, and before any 
> particular color (tonal) model is applied.  I'd imagine a 
> combination of these two techniques with a superior blender 
> like enblend should yield very high dynamic range panoramas, 
> without the need for multi-exposure blending (unless of 
> course there is a large amount of contrast in the field of 
> view of single images).  Anyone have thoughts on this?
> 
> JD
>