Vision Based Page Segmentation Algorithm: Extended and Perceived Success M. Elgin Akpınar 1 and Yeliz Yes ¸ilada 2 1 Middle East Technical University Ankara, Turkey 2 Middle East Technical University Northern Cyprus Campus Mersin 10, Turkey {elgin.akpinar,yyeliz}@metu.edu.tr Abstract. Web pages consist of different visual segments, serving different pur- poses. Typical structural segments are header, right or left columns and main con- tent. Segments can also have nested structure which means some segments may include other segments. Understanding these segments is important in properly displaying web pages for small screen devices and in alternative forms such as au- dio for screen reader users. There exist different techniques in identifying visual segments in a web page. One successful approach is Vision Based Segmentation Algorithm (VIPS Algorithm) which uses both the underlying source code and also the visual rendering of a web page. However, there are some limitations of this approach and this paper explains how we have extended and improved VIPS and built it in Java. We have also conducted some online user evaluations to inves- tigate how people perceive the success of the segmentation approach and in which granularity they prefer to see a web page segmented. This paper presents the preliminary results which show that, people perceive segmentation with higher granularity as better segmentation regardless of the web page complexity. Keywords: Web Accessibility, Web Page Segmentation, Reverse Engineering, User Study. 1 Introduction Web pages are typically designed for visual interaction. They include many visual el- ements such as header, footer, menu, etc that guide the reader. One can easily look at the visual rendering and can differentiate the segments which typically differs in back- ground color, font styles, borders or margins around the segments. On the other hand, the underlying source code typically does not provide such kind of clear segmentation or pattern. Therefore, when web pages are automatically processed by assistive tech- nologies or adapted for mobile devices this kind of information is not available. In order to address this problem, there exist different techniques for identifying vi- sual segments in a web page [3,1,7]. These approaches differ in the techniques they use for automating the process, some use machine learning techniques, some uses heuris- tics, etc. Once a web page is automatically segmented, web pages are then typically Q.Z. Sheng and J. Kjeldskov (Eds.): ICWE 2013 Workshops, LNCS 8295, pp. 238–252, 2013. c  Springer International Publishing Switzerland 2013