We propose a framework for automated classification between normal and abnormal biopsy samples of childhood brain tumour with emphasis on childhood medulloblastoma, a most common childhood brain tumour, using texture features. Texture is a measure to analyze the variation of intensity of surface of an image and the connection of pixels satisfying a repeated grey level property. The feature set consisted of a total of 172 features belonging to five texture features, GLCM, GRLN, HOG, Tamura and LBP. The performance of each feature set was evaluated both individually and in group, using six different classifiers, Linear Discriminant, Quadratic Discriminant, Logistic Regression, Support Vector Machine and K-Nearest Neighbour algorithms. Here, feature of tamura, global low order histogram and local second order GLCM outperforms the local texture measure of LBP and GRLN. Using 80 normal and malignant images of 10x magnification we obtained an optimal accuracy of 100% by combining all five textural features
