On extending the ADMM algorithm to the quaternion algebra setting

Abstract

Many image and signal processing problems benefit from quaternion based models, due to their property of processing different features simultaneously. Recently the quaternion algebra model has been combined with the dictionary learning and sparse representation models. This led to solving versatile optimization problems over the quaternion algebra. Since the quaternions form a noncommutative algebra, calculation of the gradient of the quaternion objective function is usually fairly complex. This paper aims to present a generalization of the augmented directional method of multipliers over the quaternion algebra, while employing the results from the recently introduced generalized HR (GHR) calculus. Furthermore, we consider the convex optimization problems of real functions of quaternion variable.

Publication
IEICE PROCEEDING SERIES
Srđan Lazendić
Doctoral researcher

My current research interests focus on Clifford algebra methods for efficient multidimensional data analysis and image processing. I am also interested in complex analysis, in particular Blaschke products and their properties.