A18 – Noise models and basic image restoration

In this activity this is what I did:

Procedure:

1. Add noise to the original image:

Original Image

Noises:
I applied the following noises using genradn and imnoise in Scilab and obtained their corresponding PDF p(z):



2. Filter the noisy images

I made and then applied the following filters on each type of noise in (1).

Assuming a rectangular sub-image with of size n*m, n=m=3( for nearest neighbor perimeter)

1.ARITHEMETIC FILTER

2.GEOMETRIC FILTER

3.HARMONIC FILTER

4.CONTRAHARMNIC FILTER

Q=+1,- 1




3. Determined which type of filter suited for each type noise.


In the following images the arrangement is as follows:

from the top- left-right :
(1) NOISY image (2) Filtered Image with Arithmetic Filter, (3) Arithmetic Filter
bottom-left-right is :
(4) Harmonic Filter, (5) Contraharmonic Filter : Q=-1, (6) Contraharmonic Filter : Q=+1.

1. GAUSSIAN OR NORMAL NOISE, PDF AND FILTERING

All but the geometric filter was able to somehow remove the gaussian noise in the image.

2. RAYLEIGH NOISE , PDF AND FILTERING

From these results, all the filter worked but I think the contraharmonic filter with positive Q =+1 was most effective in removing the uniform noise since the filter at Q=+1 is beginning to restore the white circle at the center.

3.EXPONENTIAL NOISE, PDF AND FILTERING

All the filter worked except the arithmetic filter.

4.GAMMA NOISE AND FILTERING


The arithmetic and geometric filter cannot remove gamma noise.

5.UNIFORM NOISE , PDF AND FILTERING

The contra-harmonic filter was most effective since its beginning to restore the white circle at the center.

6.SALT AND PEPPER NOISE AND FILTERING

Again , the contra-harmonic filter wins.

7.SPECKLE NOISE AND FILTERING

Again , the contra-harmonic filter wins (again).

SOURCE: A18 – Noise models and basic image restoration manual

ASSESSMENT: 10/10 because I was able to code and implement all the image noise and image filters :) and use imnoise. The main problem was that for the results I only used the imwrite outputs. The imshow rendered different effects that seemed better than end results saved in imwrite :) ..odd

GRATITUDE: I would like to thank Mandee and Alva for discussing with me :) (salamat) and Gibert for the rayleigh.