Reviewer #1: This is a very interesting paper and has some new results. And I recommend publication of this paper.
Reviewer #2: In this paper, the authors proposed a point-pattern matching method based on graphical model. They tried to exploit "globally rigid" to achieve 2D points set matching problem in lower complexity both in computation burden and memory requirement. They declared their method the advantage of handling occlusions.
1. The key feature for graphical model point pattern matching is that it can cast the point pattern matching problem to the inference process in graphical model. However, the algorithm presented in this paper does not use the graphical model theory and only exploit "globally rigid". In my opinion, it is not quite appropriate to take the proposed approach as graphical model approach. Therefore, the comparisons between the proposed method and that of graphical models are not totally meaningful.
2. The effects on matching results of some parameters, for example, occlusion cost (discussed in section 3.1) and K in KD-tree, should be discussed. The way to choose the parameters should be mentioned for real applications.
3. For each point in template graph, the algorithm (line 5-11 in algorithm 1) finds the nearest point in target graph. But how the KD-tree search algorithm can be used in the matching process is not mentioned there.
4. The authors have not discussed how to choose the root nodes in algorithm 1, while the last cost value depends on the root nodes selection.
5. The parameter values in Experiment section 4.2 should be given.
6. Fig. 1 would be better if it contains models of [15] [16].