The analysis of Magnetic Resonance Image has an important role in definite detection of Brain Tumor. The shape, location and size of tumor are examined by Radiology specialist to diagnose and plan treatment. In the intense work pace, it's not possible to get results quickly. At this scheme, unnoticed information can be recovered by an image processing algorithm. In this study, at database images which are collected from REMBRANT were cleared from noise, transformed with Karhunen Loeve Transform to gray level and segmented with Pott's Markov Random Field Model. This hybrid algorithm minimizes the data loss, contrast and noise problems. After segmentation stage, shape and statistical analysis are performed to get feature vector about Region of Interest. The images are classified as existing tumor or not existing tumor. The algorithm can recognize the presence of tumor with 100% and tumor's area with 95% accuracy. The results are reported to help the specialists.