In robust tumor recognition engine implementation, it is important to superimpose different types of MR images to verify the adequacy of treatment. Usually, the tumors in time- staggered MR images may vary in shape, format, orientation, angle, translation, scale and by a variety of other distortions. It is already known that various image registration techniques such as Affine transform suffers from lack of speed. Thus, the ability to extract distortion- invariant image features is highly desirable for improved speed and efficiency. We propose to explore distortion- invariant metadata extraction and subsequent classification of tumor in brain MR images.