A Fast JPEG Image Retrieval Method
Jinzhao Wu · Jisuanji gongcheng · 2005
A fast novel JPEG image retrieval method based on NMI invariable feature of DCT blocks is presented . The method directly computes the mean and variance of DCT blocks for block classification without full decompression so that it saves a lot of computation cost, and then derives a binary index map from each class of DCT blocks and extracts NMI feature vector to image retrieval. The proposed method is robust to the translation, rotation and scale transform well. Experimental results also demonstrate the method has a good performance both in retrieval efficiency and effectiveness.