An automatic rotation invariant recognition technique for colour objects and patterns

Abstract

A simple, very reliable and effective technique has been proposed for recognition of colour objects and patterns. The technique is invariant under translation, scaling and rotation, and is very useful for real-time industrial applications. The colour objects are recognized on the basis that they must satisfy both the colour and shape requirements, orientation and alignment are not critical at all. The features of the object are extracted and processed with Fourier transformation. The transformed image, together with a mapping from Cartisian Plane to the Logarithmic Plane provides a suitable transformation which removes translation, scaling and rotational dependencies. The mapped image is finally processed by using Laplacian-of-the-Gaussian filter and Log-Polar mapping. No reference points are required for processing and the image can be placed on any colour background.