eBay’s selection, value, and convenience are key components of our success. However, for many types of items, shopping online is often hindered by the inability to touch and feel the product. A high quality image can portray the product more accurately than a text description and improve the buyer’s sense of what they are purchasing. High quality images also improve the aesthetic quality of the site and can highlight features that would otherwise be hard to describe. For these reasons and more, improving the quality of our sellers’ images and the way we use image information are important topics for eBay.

Last year we implemented “More Like This by ImageSearch”, which allows buyers in our Fashion category to find items that are visually similar to other items they like. But we’ve been continuing to invest in this area. We’ve built a world-class team to develop our image processing and analysis tools. And we just received word that a poster by this team has been accepted to the WWW 2011 conference. The paper, “A Study on the Impact of Product Images on User Clicks for Online Shopping” by Anjan Goswami, Naren Chittar and Sung H. Chung, describes one of the many exciting projects that we are working on in eBay’s image analysis/computer vision team. In the paper, we describe experiments to try to augment our search ranking function with features derived from the images of an item. Some examples of such features are the amount of illumination, contrast, dynamic range, characteristics of the background etc. Our results show that better images result in better clicks and sales.

We’ll be presenting the poster in Hyderabad in March. Here is the abstract:

In this paper, we study the importance of image based features in the context of a large scale product search engine. Typically, product search engines use text based features in their ranking functions. We present an idea of using image based features, common in the photography literature, in addition to text based features. We used a stochastic gradient boosting based regression model to learn relationships between features and click through rate (CTR). Our results indicate statistically significant correlations between the image features and CTR. We also see improvements in NDCG and mean square error for CTR regression.

Stay tuned for more information about other ways we’re using eBay’s massive repository of image data and the expertise of a great team of technologists to improve the experience for our users.