ACM for segmenting intensity inhomogeneity images corrupted by a mixture of AGMG noise

Abstract

In practice, active contour models (ACMs) are effective in segmenting images, which are mostly applied in computer vision and medical imaging. However, the result may be challenged when the images contain noise and intensity inhomogeneity. Images are commonly deteriorated by a mixture of noise involving additive Gaussian and multiplicative Gamma (AGMG). To the best of our knowledge, this is the first study to formulate an ACM for segmenting intensity inhomogeneity images corrupted by AGMG noise. Our proposed mathematical model was developed using a combination of dual denoising terms to recover the noisy image and Laplacian of Gaussian (LoG) terms to smooth the targeted object. The model is solved using calculus of variations and finite difference methods. The performance of the suggested model for segmenting images was evaluated in MATLAB software and compared with current models using low and severe AGMG noise levels. Quantitative evaluations demonstrated that the proposed model outperformed existing models, achieving an average Jaccard similarity coefficient (JSC) improvement of roughly 2.5% and a 16.00% reduction in average relative error. Consequently, the proposed model proves significantly more effective for the segmentation of images degraded by intensity inhomogeneity and mixed AGMG noise.