A method that is based on Statistical mechanics has been proposed for analyzing the relationship between pixel gray values and the micro information in an image.After simulating and curve fitting of the experiment data,the image-processing model between average gray value and thresholds has been constructed using Boltzmann statistics.In order to verify the effectiveness of the model,some experiments have been carried out,in an eye-gaze input human computer interaction(HCI) system,the experiment results show that this method can achieve better results as long as the main feature of objects in the image remains the same.In these experiments,thresholds could be automatically gotten in terms of the adaptive threshold model,when the average gray values of images were between 110 and 230,the feature extracting accuracy has been 95%.