Rotation, Scale and Translation invariant image retrieval method based on Circular Segmentation and Color Density
We propose a fast and efficient method for Content Based Image Retrieval (CBIR) which uses color densities within concentric circular zones of the image, encompassing edge-pixels. This method is invariant to Rotation, Scale and Translation (RST). Small-sized feature vectors are used to store and effectively characterize the color content of the image. Consequently the memory and time required for data querying are reduced. This computationally inexpensive method is suited for portable applications. We briefly present an example of application in a handheld pictogram recognition device, used for rehabilitation and education, in which the proposed method is used as pre-selection stage of a heavier method for reducing complexity while keeping recognition accuracy.