Color detection using SRM algorithm

Runtime color detection is a difficult task, especially using low cost cameras like most people do on their computers. Brightness variability is one of the biggest problems to solve: it affects colors in the reality making them look very different from their “original” aspects.

The human eye is very powerful about recognizing areas of homogeneous colors: looking at this blue ball, we are able to say that the entire ball is of the same blue, even if each side of the ball itself looks really different in its color components: some regions appear bright, some others are shadowed, and so on. A statistical region merging algorithm tries to perform the same task of our eyes. Applied to a ball like the one above, it should decide that the entire ball is of the same “mean blue” color.

This is a movie registered from my desktop showing a color segmentation algorithm in action. In particular, we choose a SRM (Statistica region merging) implementation, which greatly helped us detecting the “hot spot” to follow into the video stream.

Explore posts in the same categories: Augmented Reality, Calameda, Java