Thread Subject: Fast way to blur a matrix?

I start with a image. It?s divided into blocks. I calculate the average of each block. In another matrix each block of the same area and position is filled with the value of the average.
 So i get a matrix comprised of blocks, with the same value within a block. Scimilar to pixilated image.
 I want to blur that matrix to smooth the edges between blocks
I am doing this using a filter, comprised of a convuluted gaussian, the problem i that it is a bit slow.
Is there a faster way to blur a image?

tks

"Ironic Prata" <lixodoironic@hotmail.com> wrote in message <hea7nk$pug$1@fred.mathworks.com>...
> I start with a image. It?s divided into blocks. I calculate the average of each block. In another matrix each block of the same area and position is filled with the value of the average.
> So i get a matrix comprised of blocks, with the same value within a block. Scimilar to pixilated image.
> I want to blur that matrix to smooth the edges between blocks
> I am doing this using a filter, comprised of a convuluted gaussian, the problem i that it is a bit slow.
> Is there a faster way to blur a image?
>
> tks

Could consider the Fourier domain, and then filtering out high frequency components. Fourier domain also can be used for the convolution and it is faster than the convolution operator.

Dear Ironic Prata!

> I start with a image. It?s divided into blocks. I calculate the average of each block. In another matrix each block of the same area and position is filled with the value of the average.
> So i get a matrix comprised of blocks, with the same value within a block. Scimilar to pixilated image.
> I want to blur that matrix to smooth the edges between blocks
> I am doing this using a filter, comprised of a convuluted gaussian, the problem i that it is a bit slow.
> Is there a faster way to blur a image?

You can use your already working block averaging with smaller blocks. A moving average would do the same (there are some tools for that in the File Exchange).
Good luck, Jan

"Ironic Prata" <lixodoironic@hotmail.com> wrote in message <hea7nk$pug$1@fred.mathworks.com>...
> I start with a image. It?s divided into blocks. I calculate the average of each block. In another matrix each block of the same area and position is filled with the value of the average.
> So i get a matrix comprised of blocks, with the same value within a block. Scimilar to pixilated image.
> I want to blur that matrix to smooth the edges between blocks
> I am doing this using a filter, comprised of a convuluted gaussian, the problem i that it is a bit slow.
========

If you insist on using a gaussian for the blur function, make sure that you are exploiting the tensorial nature of the Gaussian, i.e. make sure you filter first along the x-axis, then along the y-axis. Breaking the filtering into 1D operations will make it faster.

As pointed out, it could help to do the filtering in Fourier space, but only if the Gaussian kernel is sufficiently wide. A narrow, truncated Gaussian is more efficiently handled by direct convolution in the non-Fourier domain.

tks guys. will look in to it.

On Nov 22, 11:18 am, "Ironic Prata" <lixodoiro...@hotmail.com> wrote:
> tks guys. will look in to it.
--------------------------------------------------------------------------------------------------
In the spatial domain, look up the function conv2() and imfilter() in
the image processing toolbox. If you have large images and kernels,
then you can look into doing it in the Fourier domain.