Robust Image Segmentation Based on Superpixels and Gauss-Markov Measure Fields

Image segmentation is one of the most fundamental tasks in computer vision and image processing systems. In this work we present a multipurpose image segmentation algorithm based on SLIC-Superpixels and Gauss-Markov Measure Fields (GMMF). In the literature, both methods have shown their advantages in execution time and accuracy; however, GMMF can often blur or distort the edges of the objects in the scene, whereas SLIC is not designed for the segmentation of large, non-connected regions. This encourages combining them for a better segmentation method that is very robust to edges delocalization and has multiple applications. The proposed algorithm is able to deal with multi-channel images, different types of noise, and can be easily extended to 3D images. An experimental evaluation of the proposed method was performed using both synthetic images (with different types and levels of noise) as well as Magnetic Resonance Images for the detection of Multiple Sclerosis lesions. Preliminary results have shown robustness to noise, edge preservation and high performance for both types of applications.