Iterative Self-Improvement of Vision Language Models for Image Scoring and Self-Explanation