An active contour model driven by adaptive local pre-fitting energy function based on Jeffreys divergence for image segmentation

Active contour model (ACM) has been a competitive tool in image segmentation because of its desired segmentation result and accuracy. Nevertheless, it may become unstable while handling images with uneven intensity, different initial contours and noise. In addition, the computation process of the majority of existing ACMs is complex, which makes it time-consuming and less efficient. To resolve above issues, this paper puts forward an active contour approach driven by adaptive local pre-fitting energy function based on Jeffreys divergence (APFJD) for image segmentation. Although the computation process of the proposed model is also complex, the authors design pre-fitting functions that are computed ahead of iteration process, which reduces a great amount of computation time and increases segmentation accuracy. The key idea of utilizing pre-fitting energy is to firstly select a local region at a specific point of the entire image region and compute its median intensity. Next, this local region is grouped into two sub-regions based on this pre-computed median intensity. Afterwards, the mean intensities of these two sub-regions are calculated by averaging their image intensities respectively. Lastly, the same process is repeated at next point until all points in the whole image region are computed. After that, these two pre-fitting functions will be incorporated into a Jeffreys divergence model to construct the proposed energy function. To further improve system stability and robustness, a regularization function is defined to optimize and normalize the data-driven term and level set function. Compared with local region-based models and recently developed models, the proposed APFJD model not only greatly decreases computation cost, but also improves segmentation accuracy. Experimental results also confirm that this model is robust to initial contours with different positions as well as Gaussian noise, and effectively segments images with unevenly distributed intensity.