Abstract: Image retrieval is an important branch of computer vision. Its main purpose is to find the similar semantic images to the query image from the image database. The traditional image retrieval method is a “point-to-point” retrieval between the query image and the database. However, a single query image contains fewer category hints, that is, the category information is weaker, and the retrieval results are not satisfactory. In this paper, the "point-to-flat" category-based retrieval strategy is proposed to extend an image (point) to an image category (flat), which means the semantic extension from the individual query image to the whole image category. The proposed method mines the category information of the query image. The performance of the proposed method is evaluated on two commonly used databases. The experimental results demonstrate that the proposed method can significantly improve the performance of image retrieval.