A Cognitive Evaluation Benchmark of Image Reasoning and Description for Large Vision-Language Models