Next: Noise Estimation and Removal Up: Convolution & Deconvolution Previous: Getting started   Contents

# Practical

Load the house.l.aiff image from the images directory.

Apply the gauss function in the image calculator to the image. Remember you have to choose a suitable width using the appropriate parameters in imcalc params (Hint: use the help button to identify the parameters that gauss uses). Use a prof graph to view the affect on image structure (such as edges) and the hist and noise functions to see how it affects image noise. Repeat this for various scales of smoothing starting at a kernal width (sigma) of 1 and increasing to 5 (you must also remember to adjust the range parameter to accomodate this larger filter). Don't make the kernel too big as the execution time can increase dramatically.

Use the gauss function to enhance the edges of the image. Do this using the difference of Gaussian (DoG) approximation to the Laplacian. Can you produce a binary image showing the edge locations you have found?

Repeat the above in the frequency domain, remembering that convolution in one domain is multiplication in the other. You will need to construct an image of the Gaussian kernel of the same size as the original image (Hint: use the delta in the imcalc create tool).

Can you see any appreciable difference between the results? 4.1.

Finally attempt a Fourier deconvolution of the smoothed image (read carefully the section in the manual regarding division using "imcalc").

Attempt to answer the following questions;

• How do results compare: firstly for Fourier convolved images (at a range of scales) and secondly for spatially convolved images?

• Can you explain why this might happen?

• What are the consequences for the application of deconvolution to tasks such as image deblurring?

Next: Noise Estimation and Removal Up: Convolution & Deconvolution Previous: Getting started   Contents
root 2018-09-26