Image retrieval technology has been increasingly mature. However, the accuracy and robustness of existing methods are still limited when it comes to clothing image retrieval scenarios involving complex visual variations. To address this issue, this paper proposes a multi-level feature fusion framework and introduces multiple feature extraction and fusion modules based on this framework. With this method, we can utilize feature fusion techniques to extract and integrate features at different levels, achieving the acquisition and integration of shallow and deep features of the input image. Additionally, this paper introduces a clustering re-ranking method that combines cosine distance and the k-means ++ algorithm to calculate the spatial similarity of feature vectors and uses this to rank the image similarity. Experimental results demonstrate that the proposed framework exhibits higher accuracy and robustness in clothing image retrieval than existing image retrieval methods. Therefore, this research provides new ideas and methods for further exploration in the field of clothing image retrieval.