DIRA 2022 Poster -- Abstract Images Have Different Levels of Retrievability Per Reverse Image Search Engine
Download as PPTX, PDF0 likes5 views
This study evaluates the retrievability of natural versus abstract images using reverse image search engines, submitting 200 abstract and 199 natural images to four major search engines. Key findings reveal that Yandex excels across both image categories, while Bing and Baidu underperform; Google shows a notable retrievability gap between photographs and diagrams. Results suggest that search engines have varying definitions and effectiveness for similar-to image searches, with a general preference for natural images.
DIRA 2022 Poster -- Abstract Images Have Different Levels of Retrievability Per Reverse Image Search Engine
- 1. Abstract Images Have Different Levels of Retrievability Per Reverse Image Search Engine Shawn M. Jones smjones@lanl.gov @shawnmjones Diane Oyen doyen@lanl.gov LA-UR-22-30953 Research question When using the reverse image search capability of general web search engines, are natural images more easily discovered than abstract images? We submitted 200 abstract images and 199 natural images to four major search engines. We then applied pHash and VisHash to determine if the reverse image search engine returned the same image in its results. developed for diagrams developed for photographs Major search engines support reverse image search: upload an image as a query and receive: • pages-with results – pages containing image • similar-to results – similar images Precision@k: What percentage of images in the results are the same as the query image if we stop at k results? Retrievability: Given a query image, was it retrieved within the cutoff c? MRR: How many results, on average, across all queries, must a visitor review before finding a the same one again? Key takeaways • When they do return results, Bing and Baidu do not perform well. • Yandex performs best regardless of image category and favors natural images. • Google has a max of 54% retrievability difference between images from the categories of photograph and diagram. • Google does not perform well for similar-to results, likely indicating that their definition of similar-to differs from other search engines. • Yandex and Google consistently perform better for natural images in pages-with results. results in: green = natural images blue = abstract images