A novel stochastic resonance based closed-loop neural network method for image segmentation

Extracting multi-level information in colony images facilitates analysis and identification tasks of biomedical informatics. In order to achieve multi-level segmentation in colony images with multiple contrast levels, a closed-loop neural network model based on the stochastic resonance (SR) mechanism of neurons is proposed. First, this paper realizes the detection of transition pulses in sinusoidal, non-periodic bipolar binary signals, and one-dimensional strong and weak transition signals with multiple amplitude values. Then, through enhancement processing for the detection by the SR-based closed-loop neural network model, combined with the coati optimization algorithm, multi-target detection is achieved. Eventually, it can be applied to the segmentation of multi-level grayscale signals in two-dimensional images. Experimental results show that the proposed method can simultaneously detect strong and weak contrast edges, enrich image details, highlight image contours, and enhance the hierarchical sense of image edges, while exhibiting strong robustness against external noise. As a result, the proposed method provides a novel research framework for multi-contrast grayscale image segmentation under strong noise background.