As a competitive phenomenon that exists widely in various fields, most of the research on winner-take-all is too complicated to understand the phenomenon well. In order to explain the winner-take-all phenomenon in a simple way, this paper proposed an improved winner-take-all model, represented by discrete-time difference equations, whose state space was a matrix and the model had multi-norm selectivity, corresponding the element with the largest input value in the initial input matrix eventually wined. This paper had a rigorous theoretical analysis of the convergence and stability of the proposed model. In addition, it also applied the winner-pass model to the experiment of image processing, and compared the efficiency of the single-layer and multi-layer winner-take-all models respectively. These experiments correctly generated the winner-take-all phenomenon and demonstrated the high performance of multi-layer winner-take-all image processing. Both theory and experiment verify the correctness and effectiveness of the proposed model.

[2021-02-05] Printed Article /doc/2019.09.0639/v2