From Pixels to Feelings: Aligning MLLMs with Human Cognitive Perception of Images