Abstract
In this paper, a new DCT block-edge-pattern and high efficient method for compressed image retrieval is proposed. This method directly extracts three edge patterns from compressed image to construct 3D and 4D histogram. Then uses them and the run-length histogram as the feature -based indexing keys to retrieve images. The results of our two-stage experiments show that the superior performances of our method than other exiting method in retrieval precision, processing speed and robust to noise and color change.