Classification of Magnetic Resonance Brain Images Using Local Binary Pattern as Input to Minimal Complexity Machine

Magnetic Resonance Imaging (MRI) is a powerful visualization tool that is extensively used in medical laboratories to capture images of internal anatomy of human body. Classification of MRI brain images into tumorous and non-tumorous image is a critical and time-consuming task for the radiologist. Correct and computerized classification of MRI brain images is very important for their investigation and analysis. In this paper, we have proposed to use binary patterns (LBP) as features to classify MRI brain images into tumorous and non-tumorous. The LBP computes the relationship between central pixel and neighboring pixels of the 3 × 3 window and assigns a label to each window. The histogram of these labels is then used as a feature vector that is fed into the classification stage. The images are classified using Minimal complexity machine (MCM) algorithm. As compared to Support Vector Machine (SVM) algorithm, MCM performs better generalization and makes use of lesser number of support vectors. The performance analysis of the proposed techniques is done on the basis of accuracy calculated, and it is found that the classification rate is superior to other existing algorithms.