Prev: How to find a particular pattern in an image for a range of grayscale levels
Next: Multidimensional SAFEARRAYs via COM
From: Malcolm McLean on 28 Jul 2010 08:36
I have a large image in 24-bit rgb format (uint height x width X 3 matrix).

I want to paste a sub-image into this image.

I'd also like to paint a value into the image using a logical as a mask.

This seems impossible to vectorise.

picture(y1:y2, x1:x2, 1:3) = [255 0 0];

won't do what I want.
picture(y1:y2, x1:x2, 1) = 255;
picture(y1:y2, x1:x2, 2) = 0;
picture(y1:y2, x1:x2, 3) = 0;

works and is barely acceptable. However that only allows me to paste solid rectangular blocks of colour.

How can these manipulations be vectorised (it's also necessary to avoid making a temporary copy of the entire image)?
From: Walter Roberson on 28 Jul 2010 10:43
Malcolm McLean wrote:
> I have a large image in 24-bit rgb format (uint height x width X 3 matrix).
>
> I want to paste a sub-image into this image.
> I'd also like to paint a value into the image using a logical as a mask.
>
> This seems impossible to vectorise.
>
> picture(y1:y2, x1:x2, 1:3) = [255 0 0];

picture(y1:y2, x1:x2, 1:3) = repmat([255 0 0], y2-y1+1, x2-x1+1, 1);


> won't do what I want.
> picture(y1:y2, x1:x2, 1) = 255;
> picture(y1:y2, x1:x2, 2) = 0;
> picture(y1:y2, x1:x2, 3) = 0;
>
> works and is barely acceptable. However that only allows me to paste
> solid rectangular blocks of colour.
>
> How can these manipulations be vectorised (it's also necessary to avoid
> making a temporary copy of the entire image)?

A logical array as big as "picture" would be have the same dimensions as
"picture" but should only be 1 byte per location. Is that acceptable in
view of your need to avoid making a temporary copy of the entire image?

Is your logical mask effectively only working on a relatively small
rectangular subset of the image, with the coordinates of the subset known?
From: Malcolm McLean on 28 Jul 2010 10:55
Walter Roberson <roberson(a)hushmail.com> wrote in message
>
> Is your logical mask effectively only working on a relatively small
> rectangular subset of the image, with the coordinates of the subset known?
>
The mask represents characters in a font. I want to write a solid colour to arbitrary x, y co-ordindates within the image, using the mask. Also with option for a background.

It needs to be fast because I have lots of characters. Creating them as text objects didn't work - it took about five minutes to render, visibly slowing down as the child list became excessively long. So I'm writing them to an image buffer and blitting it out.
From: Malcolm McLean on 29 Jul 2010 04:40
Thanks. It seems that basically there isn't an answer. 3d array manipulations cannot be vectorised except in a few special cases.
From: Walter Roberson on 30 Jul 2010 12:43
Malcolm McLean wrote:
> Thanks. It seems that basically there isn't an answer. 3d array
> manipulations cannot be vectorised except in a few special cases.

Hmmm? They are as vectorizable as 2D array manipulations.
 | 
Pages: 1
Prev: How to find a particular pattern in an image for a range of grayscale levels
Next: Multidimensional SAFEARRAYs via COM