Pulse-coupled neural network (PCNN) has significant characteristics for potential high-performance image processing including image segmentation. However, segmentation accuracy is dependent on the values of network parameters. To overcome the difficulties caused by parameter settings, this paper simplified the original PCNN in terms of input and dynamic neural threshold. In the model, the generalized adjustable neural threshold is defined, and the relationship between parameters and such available information as previous output and image static properties is established. A coarse-to-fine strategy is then employed for further keeping the characteristic of the synchronous pulse, enabling the model to control the behavior of neighboring neurons. This strategy also ensures that the parameters are adjusted properly and facilitates the automatic control of the result through iteration. Finally, experiments on some synthetic and real infrared images show that the proposed model can promote segmentation capability. Furthermore, the proposed model is superior to the traditional thresholding methods and some existing PCNN-based models in terms of segmentation performance and parameter settings.