A new approach for raster images annotation

Abstract

Growth of media data creates huge problems when searching and classifying the depicted objects. Currently search is based either on surrounding context, which is irrelevant in most cases, provided image meta data sets or low or mid-level regional peculiarities of digital images. The article is focused on a specific case of image meta data - annotations: textual elements, describing or explaining objects, depicted in the image. The paper presents novel approach of data storage in the field of digital imaging, embedding annotation data in format independent way directly into digital images. The annotation scheme presented is targeted to JPEG-2000 as upcoming image storage and transmission standard, bet can be easily incorporated into other image storage formats, both lossless and lossy. The scheme allows wide taxonomy of images to be annotated: digital photos, aerial photos, scanned material, medical data sets, etc. The scheme keeps visual and structural fidelity of digital images, exploring properties of human visual system and wavelet transformation. The primary aims of the presented scheme is sharing and transfer of knowledge, like preparing material by educators, but can be extended to broader fields of application, like video, data mining or replacement for image tags and descriptions, thus creating true searchable media environment.