RE: Identifying a rectangluar area made up of small dots (80%-20% screen)

[Neil.Francis]

>-----Original Message-----
>From: Przemek Klosowski [mailto:address@hidden 
>Sent: Friday, 5 November 2010 3:06 AM
>To: address@hidden
>Subject: Re: Identifying a rectangluar area made up of small dots (80%-20% 
>screen)

>On 11/03/2010 08:27 PM, Ian Journeaux wrote:
>> Any thoughts on a technique that might be more efficient?
>What technique did you use? For a known image like you seem to have, a 
>convolution filter might work: if a is your grayscale image, and b is 
>the pixel pattern of your area, conv2(a,b) should peak where your images 
>overlap.

Perhaps, if the pixel pattern isn't regular, than it would be useful to first 
apply a texture operator of some sort, that is sensitive to irregularly spaced 
dots, but not solid colour.  Computing the variance at each pixel location, 
over a moving 5x5 window (or the dimensions of the rectangular area, if you 
know what they are), might be sufficient - the variance image would be bright 
in the dotty areas, but not as bright over solid colour - edges might also be 
bright though. 

Can't say whether this is efficient though, as this is relative to what you've 
already tried!  And what is efficient and effective on some images might be 
useless on others - it depends on your images.

Neil