An empirical evaluation on meta-image search engines

Abstract

This paper investigates the retrieval performance and necessity of major meta-image search engines (MISEs). Our study is realized in two phases. In the first phase, major image search engines (ISEs), namely, Google, Yahoo and Ask, are selected. Then, fifteen queries are determined from various topics and classified as one-, two- and three-word query groups. Each query is run on each ISE and first hundred images are considered to measure overlap ratios of ISE pairs. In the second phase, MISEs, namely, Metacca, LemmeFind, CurryGuide, iZito, and ixquick, and thirty queries are determined. Each query is run on each MISE separately and first forty images retrieved are evaluated as being "relevant" or "non-relevant". Afterwards, precision ratios of MISEs are calculated at various cut-off points for each pair of query and MISE. Furthermore, MISEs are compared with the image search engine Google in terms of precision to see the current state of MISEs. MISEs can improve the coverage by combining the results of multiple ISEs due to the overlap between ISEs is very low. Overall, iZito appears to be the best MISE. However, it is seen that MISEs, such as iZito, CurryGuide and LemmeFind, offer a good alternative for users according to ISEs. © 2008 IEEE.