Skip to Main Content
The maintenance required on modern media databases to allow efficient indexing and retrieval is becoming an ever-increasing burden. An explosion in the content of the Internet, the lowering in costs associated with imaging devices and the digital revolution have placed much higher requirements on the storage of graphics and video, both in commercial domains and at home. Fully automated indexing systems will be very desirable to manage and organise these large amounts of data. We tackle the problem of image indexing in the compressed domain using a neural network extraction technique together with a DCT domain watermark. This approach allows the indexing of the image entirely in the DCT domain, thus benefits from substantial computational savings. Coupled with DCT based watermarking, the technique allows for the real-time embedding of indexing data in popular image (JPEG) and video (MPEG) formats. Such a low-cost combination (both in processing and key storage) makes the method highly suitable for use in portable imaging devices such as digital cameras where it would be highly desirable to have some means of indexing the images they have stored.