Combined With The Level Of Intensity Inhomogeneity Image Fuzzy Set Theory Segmentation Algorithm

Image segmentation plays a critical role in Image Engineering, which to some extent determines the performance of image analysis and image understanding and hence has numerous applications including computer vision, pattern recognition, and medical image processing, etc. While much progress has been made, segmentation for image existing complex scene, especially intensity inhomogeneity, still faces many challenging problem, the stability, accuracy and speed of current segmentation algorithms also need to be improved. Hence, fuzzy theory is incorporated into level set methods to segment image with intensity inhomogeneity, which aims at improving automation, stability, accuracy and computational efficiency of segmentation. Three main works are made as follows:Firstly, local binary fitting (LBF) model and local image fitting (LBF) model with level set methods (LSM) for segmenting images with intensity inhomogeneities are sensitive to initial level set function (LSF), inappropriate initialization may cause computation burden, low stability and accuracy. LIF model with LSM for segmenting images with intensity inhomogeneities based on Fuzzy C Means (FCM) is proposed to tackle forementioned problems. The proposed algorithm first use FCM to coarsely segment images with intensity inhomogeneities and then segmentation results is used to initialize the LSF, which will be iteratively refined by LIF model with LSM. Numerous experimental results on images with intensity inhomogeneities demonstrate that:with respect to automation, stability, accuracy and computational efficiency, the proposed algorithm has salient advantages over LBF model and LIF mode with LSM.Secondly, the spatial fuzzy c means (SFCM) is first used to initially segment images with intensity inhomogeneities and initialize the LSF, which is more robust to noise and intensity inhomogeneity than classical FCM. Moreover, an improved fuzzy LSM for segmenting images with intensity inhomogeneities that incorporated with Distance Regularized Level Set Evolution(DRLSE) model is proposed, which can solve the problems of classical fuzzy LSM with low stability and numerical accuracy. Experiments conducted on images with intensity inhomogeneity demonstrate that:the improved algorithm performs favorably against the classical fuzzy LSM in regard to stability and accuracy.Finally, An improved LSM for segmenting images with intensity inhomogeneities based on bias field estimation is proposed to enhance the stability, accuracy and computational efficiency of classical LSM for segmenting images with intensity inhomogeneities based on bias field estimation. The improved algorithm is used to SFCM to coarsely segment images with intensity inhomogeneities and then uses the segmentation result to initialize LSF to avoid the problems of manual initialization. Furthermore, the improved algorithm makes use of DRLSE model to modify the energy function, which will decrease the numerical error of the improved algorithm. Extensive experiments conducted on images with intensity inhomogeneities demonstrate that the improved algorithm outperforms the classical LSM for segmenting images with intensity inhomogeneities with respect to stability, accuracy and computational efficiency.