Human factors in color-based image retrieval: An empirical study on size estimate accuracies

Abstract

In color histogram based image retrieval, a user query specification forms a histogram to be matched with those of the images in the database. The accuracy of this specification determines the matching latitude and affects directly the effectiveness of the retrieval process. The retrieval process suggests some relevant human factors, such as people's ability to estimate percentages and individual characteristics such as training, experience, and background. A two-part laboratory study was made to understand people's ability to estimate percentages of single-block and multiple-block color. The results can be applied to enhance image retrieval systems. For example, the results provide different ranges of estimates, based on statistical percentiles. Retrieval systems can use these to determine appropriate values for retrieval control parameters. The statistics can be used to help determine likely recall and precision performance. Overall, the study provides deeper understanding of people's estimate of color percentages in images. © 2003 Elsevier Inc. All rights reserved.