A Modified Hill Climbing Based Watershed Algorithm and Its Real Time FPGA Implementation
Pradipta Roy, Prabir Kumar Biswas, B.K. Das · 2016
In this paper a combined algorithm and architectural design of real time hill climbing based watershed segmentation has been proposed. We have modified the classical hill climbing algorithm in a Finite State Machine (FSM) based parallel cell network. In our approach, each cell representing a pixel propagates label of the corresponding basin to its steepest upper neighbor cells. A Locally Excitory Globally Inhibitory Oscillatory Network (LEGION) is modified for parallel label propagation among the cells. Lower complete image calculation required for overcoming plateau problem, is done by a novel minimum distance propagation method. Our design, which is implemented in VIRTEX 5 FPGA platform, achieves a speedup of the order of hundreds over the conventional ordered queue based approaches. Proposed method also outperforms most of the dedicated hardware based watershed implementations in terms of processing speed, which in our case is less than tenth of a millisecond for 512 x 512 image.