Segmentation Performance Analysis of Superpixel Methods for Brain MR Images

Magnetic resonance imaging produces large amounts of data, and the manual processing of these data results in high computational complexity. In order to solve the high computational complexity, the grouping process which is commonly used in computer vision systems is recommended. This grouping process is called superpixel. Superpixels, used espically in image and video segmentation applications, are the visiual structures composed of pixels having same color, intensity and texture behavior. Brain MR images from BrainWeb data set were studied. The performances of 3 superpixel algorithms which are widely used in the literature and have high success are compared. In order to test the success of superpixel algorithms, precise reference data of the images were used. Success metrics were calculated between the boundaries created by the superpixel algorithms and the exact reference.