Fast Convolution on Images

I am using 5X5 and 3X3 kernels to convolve with Image. As I know
convolution is computational Intensive and I have to convolve many
kernels with my image to achieve the result, the execution takes 2
seconds. I would like to reduce the processing time.

When I sequentially convolve kernel with image, the values difference
between two sequentially convolution is very less in many places,
because the pattern is not changing much. I would like to know using
this idea, can I able to reduce the computational time.

saravanan <saravananrmca@gmail.com> writes:

> I am using 5X5 and 3X3 kernels to convolve with Image. As I know
> convolution is computational Intensive and I have to convolve many
> kernels with my image to achieve the result, the execution takes 2
> seconds. I would like to reduce the processing time.
>
> When I sequentially convolve kernel with image, the values difference
> between two sequentially convolution is very less in many places,
> because the pattern is not changing much. I would like to know using
> this idea, can I able to reduce the computational time.

Hi saravanan,

If I understand you correctly, you can't. At least you can't without
introducing some type of error. 

There are at least two other possibilities, however:

  1. Use frequency-domain filtering. 

  2. Find out if your psf is separable. I really like
  Steve Eddins' article about this over at the Mathworks
  blogs:

    http://blogs.mathworks.com/steve/2006/10/04/separable-convolution/

-- 
%  Randy Yates                  % "She tells me that she likes me very much,
%% Fuquay-Varina, NC            %     but when I try to touch, she makes it
%%% 919-577-9882                %                            all too clear."
%%%% <yates@ieee.org>           %        'Yours Truly, 2095', *Time*, ELO  
http://www.digitalsignallabs.com