Semantic image retrieval basically can be viewed as a pattern recognition problem. For human, pattern recognition is inherent in herself/himself by the inference rules through a long time experience. However, for computer, on the one hand, the simulated human identification of objects is impressive at its experience (training) like a baby learns to identify objects; on the other hand, the precise identification is unreasonable because the similar features are usually shared by different objects, e.g., "an white animal like cat and dog" "a structural transportation like car and truck". In traditional approaches, disambiguate the images by eliminating irrelevant semantics does not fit in with human behavior. Accordingly, the ambiguous concepts of each image estimated throughout the collaboration of similarity function and membership function is sensible. To this end, in this paper, we propose a novel fuzzy matching technique named Fuzzy Content-Based Image Retrieval (FCBIR) that primarily contains three characteristics: 1) conceptualize image automatically, 2) identify image roughly, and 3) retrieve image efficiently. Out of human perspective, experiments reveal that our proposed approach can bring out good results effectively and efficiently in terms of image retrieval.