Eye-tracking Analysis for Automatic Documents Eye-catching Layout Retrieval Véronique EGLIN LIRIS, INSA Lyon Veronique.eglin@liris.cnrs.fr Jean CAELEN CLIPS, IMAG Grenoble Jean.caelen@imag.fr Abstract In this paper we present a synthesis of experiments of eye movement pursuit that have been applied to documents structure retrieval. The aim of this work is to propose a representation of structured documents content (the physical layout) through the simulation of a possible human inspired scan path. The research project which is presented here is based on the hypotheses that the analysis and the comprehension of real human trajectories are necessary to design a realistic automatic self- governing system that simulates eye- catching information retrieval whatever are the page designs. Keywords: Eye-tracking, scan path simulation, documents layout retrieval. 1 Cognitive approach of information retrieval The recordings of ocular movements which are measured on observers during the documents exploration give two kinds of information. Firstly, they give evidence on documents legibility: for example, they make easier the comparison between two documents which propose the same content. On the other hand, scan-path recordings underline different visual exploration strategies for a same investigation task, [10]. We have exploited two different human behaviors (inspection and skimming) and we have focused on typical strategies of structured documents reading so as to propose an image processing automated system dedicated to documents layout retrieval. The system has been designed so as to simulate human visual behavior for a global page scan. This work has been initially supported by the SHIVA 1 project and a PhD in the LIRIS, [6]. This work lies on the hypotheses that the scan- paths attest cognitive processes which are implied in the observation task, [7,8,10]. In the reading task for example, the analysis of fixations duration, their amplitudes, their locations and their variations gives evidence of mental brain operations which are automatic or under the observer control. The scan-path recording is a directly measurable sign of the observer’s care and interest. In this paper, we present the analysis of different scan-paths applied to the evaluation of Web pages design. The conclusions of the study have been supported by well known human psychovisual behaviors ([10]) to design the architecture of our system of automated layout extraction. In this paper, we do not detail any results and we report the reader to complete them with different references and bibliographies which are cited through the paper. 2 Retrieval and evaluation of documents visual content 2.1 Different exploratory situations We will not present here the eye-movements recorded device (the eye-tracker), which has been used in our experiments. We only report the reader to complementary information concerning points measures exploitation and guidance software which has have been developed by the CLIPS laboratory, see http://www-clips.imag.fr . The eye-tracker device allows recording the position of each eye of a subject during a scan on an image. With the information relative to the succession of gazing points, it's possible to study the gaze 1 SHIVA: Site Hypermédia et Inspection Visuelle Automatique (projet Emergence Rhône-Alpes)