Content-based image retrieval systems generally use multiple features to enhance retrieval performance. But how to combine multiple features is still challenging problem. To solve this problem, we extract semantic concepts from low-level features of an image and by using them the selection and combination of features are determined to achieve enhanced retrieval performance. The proposed semantics are color information including Color Coherence and Level of Detail for the color importance. To evaluate the proposed methods, experiments were performed with MPEG-7 Visual Descriptors. The number of test images is 2354. The experimental results showed that the proposed methods achieve the enhancement of retrieval performance.