A New Local-Feature Framework for Scale-Invariant Detection of Partially Occluded Objects

Detecting local features in complex images: A combination of Hough transform and moment-based approximations

“…It has been shown in earlier works (e.g. [9] and [12]) that the configuration and colour/intensities parameters of the pattern instances can be reconstructed from the values of moment computed over the circular images (both for grey-level and colour images). The parameters are solutions of certain moment-based equations individually designed for pattern.…”

“…[9], [10]) presented the method for building pattern-based approximations of circular images. Given a set of available patterns, a circular image can optimally approximated by any of these patterns.…”

“…For example, intensities (or colours) of T-junctions can be found from a system of linear equations (see [9]):…”

“…In Section II of the paper, principles of the previous algorithm (see [9] and [10]) are presented and its limitations highlighted. Section III briefly discussed how general concepts of Hough transform are tailored to the proposed method.…”

“…Thus, the local features have some visual semantics and are more informative than "traditional" keypoints which represent just locations where certain differential properties of image intensities/colours exist. Reference [9] introduced the fundamentals of approximating circular images by predefined patterns so that both the shapes and the colours/intensities within images are approximated. The approximations are built using locally computed moments of the colour/intensity functions.…”