Improved image segmentation method based on Otsu thresholding and level set techniques

Abstract Image segmentation is a complex task in the field of image processing and computer vision, as it faces challenges such as noise, low contrast, and intensity variations. Due to the simplicity, efficiency, and ease of implementation of the Otsu algorithm, as well as its ability to improve accuracy in complex scenarios and enhance robustness against noise and lighting variations, a new grayscale pixel probability selection method has been proposed based on the classic Otsu method and Chan-Vese level set. By applying the Bayesian probability formula, the probability of a grayscale pixel is determined by the global prior and non-local prior. This improved Otsu method is combined with the Chan-Vese level set curve evolution method, resulting in a better image segmentation approach. Comparative experiments have demonstrated the advantages of this method.