Scale invariant feature detection for object recognition in colour images

This paper presents a method for extracting features from color images that can be used to perform reliable matching between different views of an object or scene. We extend the concept of the feature vector given in Lowe's method to color images by defining new gradient operator or giving new feature descriptor, which result in different algorithms with different optimality in the sense of information utilization of the original color image. Generally, better algorithm optimality is coupled with higher computation complexity, which is the case in this paper. Thus, the 4 methods proposed in this paper lead to a multi-optimality and multi-complexity scheme. In practice, people can choose which method to use according to their needs and computation abilities.