Color edge detection hardware based on geometric algebra

Modern techniques treat color images as separate monochrome images for processing. Partly, because there is no straightforward generalization of linear filters available for color. However the algorithms yield more accurate results when the correlation among color bands are exploited which shows fundamental difference to process the color images. Earlier work[8] reported the transformation of the color images using Quaternion Fourier Transforms and the realization of a holistic filter based on Quaternion convolution. Here, we discuss the rotor based convolution techniques, a generalization of the previous work, within a new mathematical framework in Geometric Algebra. Based thereupon, a novel hardware architecture is proposed. Experiments show the edge detection with this technique belong to a class of linear vector filter and is holistic in nature. It is tailored for image processing applications, providing an acceptable application performance requirements. The usefulness of the introduced approach was demonstrated by analyzing and implementing a computation intensive edge detection algorithm on this hardware.