Content-based image retrieval with fuzzy clustering for feature vector normalization

In content-based image retrieval using combine multiple low-level features or deep learning features, data imbalance after normalization often occurs. Usually, the distance function is usually like the L1 or L2 norm. When the feature vector value is considered as a cluster to compute the similarity measure between feature vectors, it leads to significant deviations in the retrieval results. To ensure the data is distributed balance, we propose to use data clustering based on fuzzy clustering and then perform Gaussian normalization (feature normalization on fuzzy clustering- GFFC). Experimental results over the benchmark Corel10, Oxford5k, Paris6k datasets demonstrate the effectiveness of this propose method.