Color, shape and texture are the essential features for image and video frame processing in content-based retrieval. It is easy to obtain these features when the background of the image is simple. Once it becomes complicated color and texture, it is difficult to get such features. Two concepts, namely: relevance and focus-of-attention quoted from the computer vision can be used to locate and select the most related imformation in the image with a complicated background and the best matched image can be efficiently retrieved. In this paper we discuss these two concepts and discribe their roles in image retrieval, especially in the case of QVE (Query by Visual Example).