Naive Corners

A naive algorithm to find corners on a image.

About

When I was still studying the basics of digital image processing I came up with this "algorithm" to detect corners on a image.

How it works

Well, as the name suggests, this is a very naive approach.

Given any image we will look at it a window of 3 by 3 at a time. (We could resize the image to speed up the process, but who's in a hurry?)

3x3 window.

We will then just count how many colored pixels we have at that window. If there's 4 colored pixels, it must be a corner. In order to be considerated a valid neighbour, the color should be between some threshold. That would take care of anti-aliased borders.

Corner.

Rotated corner.

Anything else, is not a corner.

Not a corner.

Not a corner either.

Comparison

Harris Corner Detection (OpenCV)

Naives Corner Detection

Harris Corner Detection is without a doubt faster and more efficient, it also works better than mine. Those are some reasons why I called mine naive.

See it working

Run by calling

python corner.py img/mess.png

I also used this algorithm to process corners on a real time webcam video.