Region and edge based hybrid level set model for image segmentation

• Main Authors: Reddy, G. Ragotham, Shabab, Shaftain, Bhumandla, Mahesh, Banoth, Susmithakalyani, Kondra, Sreekanth
• Format: Conference Proceeding
• Language: English
• Subjects: Algorithms; Computational efficiency; Computing time; Evolution; Global optimization; Image segmentation; Topology
• ISSN: 0094-243X; 1551-7616
• DOI: 10.1063/5.0150320

The models with level set evolution contain large number of methods which are used in image segmentation with various applications. This can be categorized into two types, namely: region-level set model and edge-level set model and it has been observed that the segmentation performance is more accurate after mixing both the information i.e. edge and region. According to our knowledge gained, there are only limited numbers of theoretical functions if we merge both the region and edge function mechanisms. Generally, most of the existing hybrid models are combination of energies, which results in selecting different weight coefficients for every term and for different images in segmentation. So, to come out of these problems, there is a hybrid technique proposed which is collusion or combination of two methods, region and edge named RESLS. In this new hybrid level set model we make use of a function called normalized intensity indicator, which helps us to allow the region information to get inserted into the edge information method. Further, the energy weight coefficient terms of edge and region models can be determined by the condition of global optimization, which is formulated through the framework. We make use of different examples to demonstrate with various changes in parameters of images, either it may be density or sharpness or noise etc. The results clearly seen according to the hybrid model. Initially the original LSM was scheduled and designed by scientists Sethian and osher, which was just a simple and easy programming model but with the limitation of unbearable computing time and very low computational efficiency. However later it was updated and modified. Rather than other proposed algorithms, level set method can be more accurate, precise and effectively been used topology problems especially in the evolution of curves. It not only deals with 2D but can be further integrated to 3D images too. While implementing the level set model, we have to remember that the evolution LSM must be closed to the signed distance function. Because of its stability, irrelevancy with topology, robustness, displays a large and great number of advantages in solving the problems of corner point production, curve breaking and combining etc. Therefore there is no objection that it can be used in wide ranges.