EP3663975B1 - Method and System for Learning Pixel Visual Context from Object Characteristics to Generate Rich Semantic Images