Text this: Dynamic inputs and attraction force analysis for visual invariance and transformation estimation