Measuring capacity of cortex based image representations using spatial transformations and machine learning methods