Computer Vision and Perception #0

Spanish Version
Turns out that there are more computer vision applications around us that we actually realise. Perception as a global concept and computer vision as a specific one, are ideas really used in the real world. Behind such reality there are “simple” things like computers reading your plate number, or more “advanced” uses like computer vision aided cancer detection.

I would like to start today talking about applications of perception and computer vision technology, which will be an habitual topic in this blog. The idea here is to present and analyse problems which are currently solved with computer vision algorithms and, if it is available, I would like to describe also how the solution works. Though I am eager to start with the first application, it would be appropriate to describe very basic concepts needed to understand further explanations.

What is an image?

An image is a set of numbers! Numbers?? YES. An image is built by hundreds or even thousand of tiny squares (also known as pixels) with a number associated, where each number has assigned an intensity level. The simplest way to bear it out is thinking about a White and Black image. In that case, an intensity of 255 (the maximum value possible) corresponds to a pure white pixel, whereas a value of zero corresponds to a pure black one.

The case of color images is a bit more complicated but not too much. Firstly, we need a specific system to describe colors. There are several color models: RGB, CMYK, HSV or HSL amog others. As far as I am concerned it could be said that RGB is the most used color mode so that it will be the one I will use in this and future posts. RGB in particular, applies the same strategy that the one explained before with the Black and White Image, being the only difference that in this case there are three layers instead a single one, and each one corresponds with the intensity of an elemental color: Red (i), Green (ii), or Blue (iii). Different combinations of elemental color intensities result in different colours (there are a lot of online resources to see colors and its RGB codes i.e. ColorPicker ).

Motivation

Something that I love about robotics is that the information we use to receive from sensors doesn’t mean anything until we interpret it and give it a sense. In the particular case of computer vision,  the objective is clear: given a picture (remember, simply a bunch of numbers), being able to extract from it: shapes, textures and recognise them… For humans it is extremely easy doing this kind of tasks right?… you see a picture and that is. We have learned to do this in a completely automatic way but a computer have a couple of more difficulties and I will go through it in the foreseeable future.

There is nobody in Facebook drawing little squares around every face in a photo, or reading your plate when you go within a car park, and there isn’t anybody either within your mobile waiting you to activate in the camera the smile shot function and taking the picture in the precise moment. I hope having planted the seed of curiosity and you don’t think anymore that this kind of things are done as if by magic.

Shortly, the most typical case study. Car plate recognition.

Take Care!.

PS: I really want to finish introducing topics and going into detail, but one thing at a time.