In-camera processing of long-exposure RAW data

Started Jan 21, 2010 | Discussions thread
Yamo Junior Member • Posts: 43
Re: The Sins of the Algorithm

Thom Hogan wrote:

Anthony Medici wrote:

It sounds like as long as there are little bright points in the sky you would be happy. It doesn't matter if they are the right brightness in comparison to each other or in the right places.

Specular highlights can fall into this category if you're the right distance away from them. Indeed, that's one of the places where I first noticed the problem: in a pre-sunrise image with moonlight, I noticed that at some point the specular highlights in the water disappeared. Well, it was because they became a single pixel. This made for an artificial and strange looking transition. Those that think this is just an astronomy thing are mistaken.

The way Marianne describes the algorithm it needn't even be bright points. Virtually any texture - sand, fabric, etc. is going to have many, many single pixel local maxima with the visual impact of overall reduced micro-contrast of the texture.

Imagine the image data as a height field (a surface with each pixel value represented as height)... Now, take a hammer to all the peaks and flatten them down to their nearest neighbor (height-wise).

Take any photo (taken on the shorter side of 1/4 second) and zoom in until you see individual pixels as squares 8 to 10 screen pixels to one image pixel. Any individual brighter pixel you see would have been hammered with this algorithm. Try individual channels

And... demosaicing will spread the impact over, what, an order of magnitude more pixels? It's just so bad...


Post (hide subjects) Posted by
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow