An Enhanced Segmentation Method by Combining Super Resolution and Level Set

Abstract

High accuracy in image segmentation is highly demanded in today’s technological world. In this paper combining super resolution image reconstruction with level set image is proposed as a way to improve image segmentation. The term Super resolution, resolution enhancement, is a process to increase the resolution of an image. This improvement quality is due to sub-pixel shift of low resolution (LR) images from each other between images. In fact, each LR image has new information of the image and the main aim of super resolution is to combining these LR images to enhance the image resolution. Following this method, allows users that without any demand for additional hardware, overcoming the limitations of the imaging system. Moreover, the main goal of segmentation is to distinguish an object from background. Segmentation can do that by dividing pixels of an image into prominent image regions. In fact, a specific region is corresponding to individual objects or natural parts of objects. Segmentation can be used in different fields such as image compression and image editing. Vandewalle algorithm and level set segmentation are used in the super resolution and segmentation part respectively. Additionally the regularity of the level set function is conserved via level set regularization term to evade expensive evolving level set function re-initialization. Experiment results for real and magnetic resolution (MR) images indicate the performance of our method. Using level set segmentation technique with super resolution, improves segmentation results in both normal and MR images.