We propose a signature scheme for identifying a related class of images based upon the content of the images. With the proposed scheme, we represent an image to a collection of rules based upon a technique using relationships derived from the pixels of images. This collection of relationships or rules is called Finite Inductive sequences. These rules make up a collective storage structure which can be used to process an image. The rules used in processing an unknown image characterize the image. The storage requirement increases with the number of rules for an image, which is on the order of the number of pixels within the image. One way to alleviate the storage requirement associated with large images is to process the image by using a wavelet transform, and then considering only the resulting high frequency component of the transform as the input to this process. When a new image is submitted, the rules are used to recognize similarities between the stored image and the new image. The process will provide an interlinking mesh to images that are similar or have similar components, as a background process. Retrieval then can be done without additional work at the moment of retrieval.
展开▼
