Newsgroups: sci.image.processing
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!news.mathworks.com!udel!gatech!howland.reston.ans.net!pipex!bbc!alanr
From: alanr@rd.bbc.co.uk (Alan Roberts)
Subject: Re: Chromakeying
Message-ID: <D3owBA.JEG@bbc.co.uk>
Sender: usenet@bbc.co.uk
Nntp-Posting-Host: sunf0.rd.bbc.co.uk
Organization: British Broadcasting Corporation, UK
X-Newsreader: TIN [version 1.2 PL0]
References: <ari.792203452@Canine>
Date: Wed, 8 Feb 1995 16:25:57 GMT
Lines: 57

Ari Schindler (ari@Canine.CAD.UCLA.EDU) wrote:
:   I've been playing with image processing lately, and a number of questions
: have come up. I was hoping that someone out in netland might have the answers:
: 1. You've got an image of a subject standing in front of a blue screen. What
: are some good functions for generating a mask that includes only the blue
: portion? "Good" here means flexible, and simple enough that it can be
: calculated quickly in software.

In TV systems, this usually means looking not for parts of the signal that
are maximally blue, but parts of the signal where there is the biggest
difference between the blue signal and luminance, ie B-Y is positive and
big.

: 2. In the same situation, you'd like to remove the "blue halo" that surrounds
: the subject (presumably due to the 'anti-aliasing' of the subject and
: background that happens in the real world - if only the real world were made
: up of properly aligned pixels...). What are some good functions for doing
: this (same sort of "good" as in previous question)?

The blue halo is a function of the blue background, which is usuall quite
a light blue in order to keep the camera noise down. Light from the blue
background spills onto the foreground subject, and is most visible at
the edges.

: 3. I've noticed some interesting artifacts in many of the images that I've
: looked at. The most prevalent is a sort of short, wildly-miscolored stripe,
: about 1 pixel wide, that I find in areas of an image where color is changing
: radically from pixel to pixel. The classic case is that in an image in which
: a model's hair is blowing in the breeze, there will be countless little 'hash
: marks' (they look like they might be hue-shifted into neon green or red from
: the local colors) at the points where hairs cross each other.
:   I suspect that this is some artifact of the interpolation that 1CCD cameras
: do - am I correct?

Not a bad conjecture. In TV terms, the switching is not done pixel by
pixel because the camera noise would make ragged edges. It is usual to
spatially filter the key signal and use it as a multiplier rather than
as a hard swith. The effect you report could well be the result of using
a key signal as a hard switch.

: 4. I've also noticed some peculiar horizontal stripes in the captures from
: a JVC TK1070 1CCD camera. They cover the whole image, and look rather obnoxious.Any idea what this might be?

Pass.

:   Any and all help would be greatly appreciated - many thanks in advance!
: BTW, my access to this group is a bit intermittant, so responses via email
: would really help me out.

Hope that helps.

--
************* Alan Roberts **************
* BBC Research & Development Department *
* My views, not necessarily Auntie's    *
*    but they might be, you never know. *
*****************************************
