PhotoIndex: Image retrieval different from cbir and tagging

A different approach than Content Based Image Retrieval (CBIR) and metatagging
The PhotoIndex project is an approach to storage and retrieval of digital images. The idea is to trace a photo and store the image's abstract representation. PhotoIndex's target users could be a selection of consumers and/or professionals such as photography enthousiasts, graphic designers and multi media developers. They are aware of composition and colour within the images they create or look for.

PhotoIndex is about defining meaningful image spaces. It uses CBIR's advantage of abstraction by using geometrical shapes, simple patterns and a limited set of colours. Combined with graphical archetypes (symbols) that resemble 'person' or 'water' for example, the advantages of manually entered discriptive metadata's semantics are brought back, to a satisfying level.

Metadata and tagging versus Content based image retrieval

Currently, there are two main approaches for digital image storage and retrieval. One is to use manually entered discriptive metadata (or tags), which is about textual descriptions of images. The other is Computer Vision and Content-Based Image Retrieval (CBIR). CBIR indexes images by using feature extraction like colour, structure, etc. by means of mathematical algorithms.

Metatagging

Creating manually entered descriptive metadata (tagging) is a tedious and time consuming process, prone to discrepancies because of type errors, differences in languages, cultural interpretation and selection of words.

The problem with tagging is that the approach is communicating about the visual domain by using the textual domain. The reason for this is mainly because computers can not compare images by contents. If humans help computers by describing images, the computer can compare the descriptions, rather than the actual images.

Content-based image retrieval

With Content-based Image Retrieval, images can be indexed relatively quickly and easily. However, the many image feature extractions are hard to program/develop and hard to understand for end users. Also, lack of semantics very often gives bad results after querying for an image from a Content Based Image Retrieval system.