Algorithmic foundations for hardware implementation of scale-insensitive MSER features
Andrzej Stefan Sluzek, Hani H. Saleh · 2016
The paper discusses algorithmic aspects of SIMSER feature detectors (SIMSERs are an efficient, multi-scale generalization of MSER features) targeting the prospective hardware implementation. In particular, it is shown that the architecture for SIMSER detection can be built over the existing architecture of MSER detection with a few modules added. The additional modules are: (1) image smoothing filter to generate the image in a sequence of scales, (2) replicas of the data memory (to handle results computed in a few neighboring scales) and (3) a module to combine data from the neighboring scales. The paper specifies the algorithmic structure of those additional modules, so that inexpensive hardware implementation of SIMSER detector on top of the exiting MSER hardware architecture becomes feasible.