Over seventy percent of Web users connect to the Web with a 28.8KBps (or slower) modem (Pitkow and Kehoe, 1996). This statistic along with need for optimizing image database retrieval systems are impetus for our research. This paper presents an attempt to find an optimal image size for thumbnails, maximizing recognition while minimizing overall file size (which takes into account number of colors, the height, and the width). The paper details an experiment designed to ferret out the optimal image thumbnail size. In the experiment subjects are shown twenty images at five randomly varying sizes, then they press a button and then are asked to describe to image they saw. The reaction speed is measured and their qualitative response is judged for accuracy. An average level of recognition and click time, will be measured for each image size. This information will provide baseline for the recommended "optimal" thumbnail size. We believe that if smaller thumbnails provide similar recognition rates to larger thumbnails, designers of image archive systems (particularly web-based systems) will benefit from the findings of this paper.
Of particular interest to us is the image the subjects accurately responded to the quickest. This is the subject of few studies to date, yet is the concern of many in this field. If the archive systems store less data or transmit less bandwidth, and get similar results, the World Wide Web might experience less net traffic if standards based on our findings are adopted.
In this survey, five treatments (pictures at different sizes) are used across the various pictures. The subjects' ability to discern an image from a reduced size thumbnail is measured both quantitatively and qualitatively. The quanititative measure is the percentage of images a subject correctly identifies at each image size. The qualitative measure is the percentage of images at each image size that the subject believes to have correctly identified.
In ideal conditions, it would be best to have a subject group that demographically represents the WWW user population, however, we tested mainly WWW-using college students at the University of Maryland. People with vision problems were not excluded from participating. The subjects were be asked to use their corrective devices (glasses or contacts) during the testing. All subjects who required required corrective devices used them during the experiment.
Next Section: Experiment