PhotoSqueak 1.0

by Juan Manuel Vuletich

Part 3

Exercise 1

Introduction

Histogram Equalization.

Development

First I implemented Histogram equalization of monochromatic images. The contrast is greatly enhanced, and no artificial artifacts are seen.

Then I tried the technique to color (multiband) images. I equalized each band on its own. The result is not good, because the colors in the image are modified. I also included the posibility to equalize the norm of the pixels, weighting together all the color bands. The result is much better. The image on the left is the norm-equalized one, and the image on the righ is the band-equalized one.

Exercise 2

Introduction

It was asked to do a local minimum image classification.

Development

I implemented a "smooth histogram". Local minimum are found, and the image is classified. I did not implement this techique on color (multiband) image. This would require finding minimum values on a multi-dimensional space, and thus requires further study.

Exercise 3

Introduction

It was asked to generate noise images with several distributions.

Development

I first implemented a normal distribution noise generation, based on the Box-Müller method. Then I implemented the noise image generation. For multiband images, each band is generated separatedly. When trying to show histograms it became evident the need for aproximate histograms, because in noisy images, no pixel value is usually used twice. We would end with a huge histogram where most entries have only one or two pixels. What we really need is the "density" of the histogram, in pixel value intervals.

An Aproximate Histogram is a hibrid between an static and a dynamic histogram and is really useful for Float images. The actual range of the values of the image is taken and divided in a number of equal width intervals. For each pixel value in the image, the corresponding interval is found. So, the histogram counts the pixels in each interval. See how it looks in this example:

Exercise 4

Introduction

It was asked to contaminate images with Salt & Pepper, Aditive and Multiplicative noise, and in different distributions, the test images. It was asked to apply different filters and to draw conclusions. I didn't include examples here, but you can try them if you want.