Towards Interpreting Visual Information Processing in Vision-Language Models