Psychophysics of Autostereogram Videos: Blur, Contrast and Repetition Period Georgios Papadimitriou I V N E R U S E I T H Y T O H F G E R D I N B U Master of Science in Artificial Intelligence School of Informatics University of Edinburgh 2010 Abstract Autostereograms are single image stereograms that take advantage of the binocular fusion and stereopsis of the human vision system. In this way, through autostere- ograms, we visualise in three dimensions objects or scenes that are embedded in two- dimensional images. In addition to (static) autostereograms there are autostereogram videos which are either videos that are created from animated depth masks of ob- jects/scenes or videos that are composed of sequences (frames) of static autostere- ograms. In the work presented in this thesis, we investigated the psychophysical as- pects of Random Dot Autostereogram (RDA) videos with respect to blur, contrast and repetition period of the random dots that constitute the repetitive patches inside a ran- dom dot autostereogram. The approach we followed focused on human performance data gathering by conducting experiments on human subjects that we tested for stere- opsis achievement and how fast it (stereopsis) was achieved. The stimuli we used were autostereogram videos of basic objects (cubes, tubes, pyramids, disks and pentagons) in which we varied the setting of one of the aforementioned features (blur, contrast etc.) each time while keeping the rest fixed. With respect to blur, our findings showed that there is an upper threshold of uniform blur radius at 33-35 pixels above which subjects were unable to achieve stereopsis. In addition, we found a threshold at 0.02 contrast below which subjects were also unable to achieve stereopsis. Regarding repetition pe- riod, we found that there is an optimal range of settings (70-100 pixels) for repetition period within which subjects identified the objects inside the autostereogram videos faster and a range (30-100 pixels) outside which misidentification of objects and lack of 3-D perception are present. Our findings, in the vast majority of the feature settings tested, showed no statistically significant differences in performance between males and females and between people that wear glasses or contact lenses and people that do not. On the contrary, we found statistically significant differences in the performance of experienced (in watching autostereograms) subjects when compared to inexperi- enced ones, with experienced subjects performing better. i Acknowledgements First of all, I would like to thank my supervisor, Robert Fisher, for his invaluable guid- ance and support throughout the project period. I would also like to thank Dr. Richard Shillcock for the help he provided in analysing the data. Many thanks to my family that supported me both financially and emotionally during this project and throughout the whole year that I spent in Edinburgh. I would also like to thank all my friends here in Edinburgh for all the great times we had together. Last but not least, I would like to thank my girlfriend, Stella, for everything she has done for me the past year and especially during writing this thesis. ii Declaration I declare that this thesis was composed by myself, that the work contained herein is my own except where explicitly stated otherwise in the text, and that this work has not been submitted for any other degree or professional qualification except as specified. (Georgios Papadimitriou) iii To my family. iv " Vision is the Art of Seeing what is Invisible to Others " (Jonathan Swift) v Table of Contents 1 Introduction 1 1.1 Motivation, Aims and Objectives . .2 1.2 Summary . .2 2 Background and Related Work 5 2.1 Background . .5 2.1.1 Types of Stereograms . .5 2.1.2 Autostereoscopic Perception and Autostereogram Observation Techniques . .9 2.1.3 Platforms used . 11 2.2 Related Work . 16 3 Creating the Autostereogram Videos 18 3.1 Autostereogram Video Creation (Steps 1 and 2) . 19 3.2 Autostereogram Video Creation (Step 3) . 21 3.2.1 Creating and Selecting Videos of Different Blur . 22 3.2.2 Creating and Selecting Videos of Different Contrast . 27 3.2.3 Creating and Selecting Videos of Different Repetition Period . 32 4 Gathering and Analysing Human Performance Data 34 4.1 Human Subjects . 34 4.2 General Experimental Procedure . 35 4.3 Stimuli . 40 5 Experimental Results and Statistical Analysis 44 5.1 Experiments with Autostereogram Videos of Different Blur Radius . 44 5.1.1 Subjects Wearing Glasses Versus Subjects not Wearing Glasses (different amounts of blur) . 48 vi 5.1.2 Female Versus Male Subjects (different amounts of blur) . 52 5.1.3 Experienced Versus Inexperienced Subjects (different amounts of blur) . 54 5.2 Experiments with Autostereogram Videos of Different Michelson Con- trast . 57 5.2.1 Subjects Wearing Glasses Versus Subjects not Wearing Glasses (different Michelson contrast settings) . 60 5.2.2 Female Versus Male Subjects (different Michelson contrast set- tings) . 62 5.2.3 Experienced Versus Inexperienced Subjects (different Michel- son contrast settings) . 64 5.3 Experiments with Autostereogram Videos of Different Repetition Periods 66 5.3.1 Subjects Wearing Glasses Versus Subjects not Wearing Glasses (different repetition periods) . 69 5.3.2 Female Versus Male Subjects (different repetition periods) . 72 5.3.3 Experienced Versus Inexperienced Subjects (different repeti- tion periods) . 74 6 Conclusions and Future Work 77 6.1 Overview . 77 6.2 Conclusions . 78 6.2.1 Conclusions for Autostereogram Videos of Different Blur . 78 6.2.2 Conclusions for Autostereogram Videos of Different Michel- son Contrasts . 79 6.2.3 Conclusions for Autostereogram Videos of Different Repeti- tion Periods . 80 6.2.4 Future Work . 81 A E-mail for Human Subject Gathering 83 B Instructions for the Experimental Procedure/Consent Form 84 C Human Performance Data 86 C.1 Blur Data . 87 C.2 Michelson Contrast Data . 97 C.3 Repetition Period Data . 107 vii Bibliography 117 viii Chapter 1 Introduction The world around us, as we perceive it, is a three-dimensional (3-D) environment in which every object can be described by its three dimensional coordinates (x, y, z). In most cases the term 3-D is misused and it refers to the techniques that people use to represent 3-D objects on 2-D planes (e.g. papers, computer screens etc.) [21]. By varying for instance an object’s relative size, the shade and the light that are used to represent the object, we can incorporate depth information in a 2-D image and perceive it as a 3-D one [21]. These are some of the techniques used in order to enable humans perceive the world in three dimensions with their vision system. In addition, human eyes can see two different but similar images which are later combined and perceived as one 3-D image [21]. This phenomenon is called binocular fusion and the sense of depth in the images as we perceive them with our eyes is called stereopsis [21]. According to [13] the sense of depth can arise from stereopsis without the use of any other technique. The difference between stereopsis and the techniques that are used lies in the fact that, when using techniques such as shading to represent a 3-D object in a 2-D plane, humans are able to see the 3-D object even with one eye, which of course is not the case in stereopsis where a combination of the two similar images provides the final image of what we see. Autostereograms (see figures 2.4, 2.5) is a type of stereograms that allows us to represent 3-D objects and scenes using ordinary display devices and means (computer screens, paper etc.) by taking advantage of the ocular fusion and stereopsis of the hu- man vision system. More specifically, autostereograms are single image stereograms through which we are able to visualize 3-D objects inside a 2-D image. So far, there is a lot of research on the space of perceptibility and psychophysics of static and dy- namic stereograms as well as their applications in health sectors [11], [12], [8], [24], 1 Chapter 1. Introduction 2 [20], [22], [9]. What has not been considered yet much is the space of perceptibility of autostereogram videos. Regarding autostereogram videos, there are parameters that affect how effectively humans perceive them or even whether they can perceive them or not. Like in static types of stereograms, factors such as image blur, the disparity between the background and the foreground (3-D object) of the image, the 3-D object represented in it, the colours used to generate an autostereogram and the repetition pe- riod of pixels inside the autostereograms (applicable for random-dot autostereograms) should also affect the outcome of an autostereogram video. The question that arises at this point is simple: which factors, how and to what extent affect the perception of an autostereogram video by a human? 1.1 Motivation, Aims and Objectives Despite the fact that stereoscopic vision was discovered almost two centuries ago, de- spite the various types of stereograms developed since, and despite the numerous stud- ies on the psychophysics and the space of perceptibility of static stereograms, there is less work on dynamic random dot stereograms and no work known to us regarding the space of perceptibility and psychophysics of autostereogram videos (the different types of stereograms will be explained in section 2.1.1). This is what motivated us to conduct research on the specific topic and set the basis for further research on this field. The goal of our research is to study the psychophysics of autostereogram videos by conducting experiments on human subjects, gathering human performance data and analysing them in order to better understand the human vision system and find the thresholds under or above which humans are not able to perceive autostereogram videos.