Robust image segmentation and bias field correction model based on image structural prior constraint

• Published in: Expert systems with applications 2024-10, Vol.251, p.123961, Article 123961
• Main Authors: Zhao, Wenqi, Sang, Jiacheng, Shu, Yonglu, Li, Dong
• Format: Article
• Language: English
• Subjects: Adaptive bias correction; Binary level set; Image segmentation; Reflectance prior constraint; Retinex
• ISSN: 0957-4174; 1873-6793
• DOI: 10.1016/j.eswa.2024.123961

In this paper, we propose an advanced variational model for image segmentation and bias correction. In contrast to the majority of existing level set segmentation models that only consider illumination bias fields, we additionally consider the impact of image reflectance on segmentation accuracy. Our method is capable of effectively segmenting images with blurry edge structures affected by non-uniform illumination. In order to enhance segmentation efficiency, we directly segment the underlying structures of the images, construct spatial prior and apply adaptive regularization constraints on the structural component. Therefore, in the process of segmentation, the proposed algorithm can accurately identify object boundaries without being affected by the environment. Besides, the GL operator is applied to enhance the robustness of the model against noise. Furthermore, we use the alternating direction method of multipliers and the operator splitting algorithm for numerical solution. The experimental results obtained from various sorts of images illustrate that our model outperforms many leading-edge level set models with regard to robustness, corrected results and accuracy. •A novel image segmentation model based on image structural prior is introduced.•An adaptive regularization constraint for adaptive restoration of image intensity.•The model is robust to images with noise and intensity inhomogeneity.•The split operator and alternating minimization algorithm are combined.