Sequential and indexed two-dimensional combinatorial template matching allowing rotations

Rotation and Lighting Invariant Template Matching

“…Table 1 shows the time complexities (under the word RAM computation model) of our algorithms for computing these distances for every possible rotation of a pattern centered at a fixed text position. We remark that a lower bound to this problem is Ω(m 3 ), and an algorithm whose time complexity matches this lower bound was given in [14,15] without lighting invariance (see Section 3 for more on lower bounds). On the other hand, in the integer case it is trivial to obtain O(σm 3 ) time by trying out each possible transposition (i.e., difference among gray levels).…”

“…Since then, several fast filters have been developed for diverse matching models [18,9,19,14,13,15]. These represent large performance improvements over the FFT-based approach.…”

“…These represent large performance improvements over the FFT-based approach. The worst-case complexity of the problem was also studied [1,14,15]. However, lighting invariance has not been considered in this scenario.…”

“…The aim of this paper is to enrich the existing algorithms for rotation invariant template matching [14,15] with the techniques developed for transposition invariance [21,10,11] so as to obtain rotation and lighting invariant template matching. It turns out that lighting invariance can be added at very little extra cost.…”

“…In [14,15], some of the models considered were useful only for binary images, a case where obviously we are not interested in in this paper. We will consider models that make sense for gray level images.…”