Content-based image retrieval and its benefits for the stock photography market

Show full item record

Title: Content-based image retrieval and its benefits for the stock photography market
Author: Padeste, Romano
Abstract: The development of powerful low-cost desktop computer systems has changed the pre-press business where tight deadlines must be met per sistently. An increasing number of newspapers and magazines are acquiring, handling, and storing images digitally while the use of hardcopies and slides decreases. Today's computers and high capacity storage-media enable stock pho tography agencies to build digital image databases, giving users fast access to large numbers of images. However, the transition from analog to digital image archives imposes new problems: with thousands of images at hand, the search for a particular image may turn into the search for the needle in a haystack. The first image Database Management Systems (DBMSs) were extended text DBMSs, which stored the image data along with a set of manually entered descriptive keywords. The major problem with this approach is that there is no generally agreed-upon language to describe images. Even sophis ticated DBMSs are unable to detect synonyms; hence, an image described with certain properties such as "curvy" may not be found if a user enters "wavy" as a search criterion. Furthermore, some image properties are hard to describe with keywords. A search is likely to fail if properties were not described at the database population stage when images are added to the database. Finally, assigning a sufficient set of keywords to every image adds a tremendous amount of labor to the population stage. Research at many scientific institutions and companies is geared towards overcoming the shortcomings of image DBMSs with keyword-based search engines. Pattern recognition which allows for comparing images based on their visual content is being introduced to image DBMSs, improving the accuracy of search engines. Sketches, sample images, and other means of describing the visual content of images may be used as search criteria in addition to keywords. This thesis project summarizes the basics of pattern recognition and its applications in image database management for contentbased image retrieval. The purpose of this thesis project is to determine the impact of contentbased image retrieval on the stock photography market in the near future. In order to obtain the necessary information, two different questionnaires were sent out to a number of selected stock photography agencies, newspapers, and magazines. The evaluation of the replies was conducted for the three groups separately. The replies from stock photography agencies showed a high interest in digital image archives. They also showed concerns about increased overhead with digital archives. The estimated amount of work required for categoriz ing images and assigning keywords ranged from fifty to ninety percent as compared to ten to fifty percent for scanning. All survey participants agreed that pattern recognition can improve the accuracy of keyword-based search engines. However, they all denied that this approach would reduce the need for assigning keywords. Different needs could be determined for newspaper and magazines. Newspapers rely heavily on keywords since images are often chosen based upon the circumstances under which they were taken while their visual con tent may be secondary. Therefore, newspapers' profits from content-based image retrieval are minute. For magazines, the visual content of images seemed to have a higher priority and the appreciation for corresponding search capabilities was accordingly higher. To summarize, users of digital image archives can profit from contentbased image retrieval if the visual content is an important issue. For image providers, there are a number of reasons that delay the transition to contentbased image retrieval. Currently, there is only one shrink-wrapped commer cial product available that meets the needs of stock photography agencies. This product requires additional work for fully exhausting its capabilities. Finally, many companies have already built their image database and the transition to another system is time-consuming, expensive, and risky.
Record URI: http://hdl.handle.net/1850/13789
Date: 1995-12

Files in this item

Files Size Format View
NRestrepoThesis10-1995.pdf 148.0Mb PDF View/Open

The following license files are associated with this item:

This item appears in the following Collection(s)

Show full item record

Search RIT DML


Advanced Search

Browse