Downsampling an image

Hi everybody,

I want to downsample a 49x49 image into 7x7 image using gaussian filter of 31x31. The alternate approach I tried (but not getting the accurate result) is as follows.

Any help is very much appreciated.

with regards, Venkat

clear all,clc
I = imresize(imread('pout.tif'),[49 49]); % I used matlab image as example of 49x49 image instead of my experiment image
hgausspymd = video.Pyramid;
hgausspymd.PyramidLevel = 3;
I_down = mat2gray(double(step(hgausspymd, I))); 
imshow(I_down)

[David Young]

Why 31x31 for the filter? This is an enormous filter to apply to a 49x49 image, and will give rise to serious problems at the image boundaries, since it will only be wholly inside the image when centred on pixels in the middle 19x19 square.

[Image Analyst]

Exactly what does "accurate" mean to you? As David says the final output image will be nearly all edge effects since very little will fit the 31x31 filter. There are a variety of ways to take edge effects into account. Which way do you prefer? Which way are you using when you define accuracy?

Answer 1

[Oliver]

a(1:2:end,1:2:end)

[Walter Roberson]

That would downsample to 25 x 25 rather than 7 x 7, and would not use a gaussian filter as asked by the original poster.

Answer 2

[Venkat R]

Dear friends, thanks for your comments.

I am using the following code.

"accurate" means, should be able to summarize the information content in 49x49 images. Not just the peak/central regions etc

h = fspecial('gaussian',[31 31],3.5);
idown= imfilter(bw4,h,'symmetric');

bw4 is a 17x17 matrix. If I take first 7 alternative values in it, it is serving me the purpose

with regards, Venkat