Eye gaze interaction enables users to interact with computers using their eyes. A wide variety of eye gaze interaction techniques have been developed to support this type of interaction. Gaze selection techniques, a class of eye gaze interaction techniques which support target selection, are the subject of this research. Researchers developing these techniques face a number of challenges. The most significant challenge is the limited accuracy of eye tracking equipment (due to the properties of the human eye). The design of gaze selection techniques is dominated by this constraint. Despite decades of research, existing techniques are still significantly less accurate than the mouse. A recently developed technique, EyePoint, represents the state of the art in gaze selection techniques. EyePoint combines gaze input with keyboard input. Evaluation results for this technique are encouraging, but accuracy is still a concern. Early trigger errors, resulting from users triggering a selection before looking at the intended target, were found to be the most commonly occurring errors for this technique. The primary goal of this research was to improve the usability of gaze selection techniques. In order to achieve this goal, novel gaze selection techniques were developed. New techniques were developed by combining elements of existing techniques in novel ways. Seven novel gaze selection techniques were developed. Three of these techniques were selected for evaluation. A software framework was developed for implementing and evaluating gaze selection techniques. This framework was used to implement the gaze selection techniques developed during this research. Implementing and evaluating all of the techniques using a common framework ensured consistency when comparing the techniques. The novel techniques which were developed were evaluated against EyePoint and the mouse using the framework. The three novel techniques evaluated were named TargetPoint, StaggerPoint and ScanPoint. TargetPoint combines motor space expansion with a visual feedback highlight whereas the StaggerPoint and TargetPoint designs explore novel approaches to target selection disambiguation. A usability evaluation of the three novel techniques alongside EyePoint and the mouse revealed some interesting trends. TargetPoint was found to be more usable and accurate than EyePoint. This novel technique also proved more popular with test participants. One aspect of TargetPoint which proved particularly popular was the visual feedback highlight, a feature which was found to be a more effective method of combating early trigger errors than existing approaches. StaggerPoint was more efficient than EyePoint, but was less effective and satisfying. ScanPoint was the least popular technique. The benefits of providing a visual feedback highlight and test participants' positive views thereof contradict views expressed in existing research regarding the usability of visual feedback. These results have implications for the design of future gaze selection techniques. A set of design principles was developed for designing new gaze selection techniques. The designers of gaze selection techniques can benefit from these design principles by applying them to their techniques

In their first year, infants’ ability to follow eye gaze to allocate attention shifts from being a response to low-level perceptual cues, to a deeper understanding of social intent. By 4 months infants look longer to uncued versus cued targets following a gaze cuing event, suggesting that infants better encode targets cued by shifts in eye gaze compared to targets not cued by eye gaze. From 6 to 9 months of age infants develop biases in face processing such that they show increased differentiation of faces within highly familiar groups (e.g., own-race) and a decreased differentiation of faces within unfamiliar or infrequently experienced groups (e.g., other-race). Although the development of cued object learning and face biases are both important social processes, they have primarily been studied independently. The current study examined whether early face processing biases for familiar compared to unfamiliar groups influences object encoding within the context of a gaze-cuing paradigm. Five- and 10-month-old infants viewed videos of adults, who varied by race and sex, shift their eye gaze towards one of two objects. The two objects were then presented side-by-side and fixation duration for the cued and uncued object was measured. Results revealed 5-month-old infants look significantly longer to uncued versus cued objects when the cuing face was a female. Additionally, 10-month-old infants displayed significantly longer looking to the uncued relative to the cued object when the cuing face was a female and from the infant’s own-race group. These findings are the first to demonstrate that perceptual narrowing based on sex and race shape infants’ use of social cues for allocating visual attention to objects in their environment.

Syftet med denna studie är att utforska den manliga och kvinnliga blicken i George R.R. Martins fantasinovell, A Game of Thrones (1996). Jag föreslår att skillnader i hur den manliga och kvinnliga blicken betraktar sitt objekt, kan avslöjas genom kritisk analys av kvinnliga huvudkaraktären Sansa Stark. Vidare menar jag att patriarkala strukturer kan synliggöras genom analys av manliga blickar som riktas mot den kvinnliga karaktären. Min analys av Martins fantasinovell har genomförts med hjälp av ett teoretiskt ramverk baserat på framförallt Laura Mulveys artikel Visual Pleasure and Narrative Cinema (1975) och Rachel S. Grates tes ”Love at First Sight? Jane Austen and the Transformative Male Gaze” (2015). Min analys visar att det finns skillnader i hur den kvinnliga och manliga blicken betraktar i sitt objekt. Den kvinnliga blicken tenderar exempelvis att vara mer mångfacetterad i sin värdering av ett objekt. Vidare visar analysen att de manliga blickarna avslöjar patriarkala strukturer.

Critics of Clarice Lispector often identify feminist themes relating to voice, gender, and the male gaze in her creative work. Lispector’s collection of short stories Laços de família demonstrates the way patriarchal society sets limits on the identity of women. Laura Mulvey’s concept of “the male gaze” provides a useful tool for understanding how men marginalize, objectify, and subordinate women through visual regimes of control, yet Mulvey’s concept does not fully encapsulate the scope of male oppression explored in Laços de família. In fact, Lispector draws upon a variety of senses and metaphors related to consumption through a mode I call food femininities to display how men consume their female counterparts in society. More specifically, Lispector’s collection Laços de família invokes, presents, and uses food, food imagery, food vocabulary, and food metaphors as a central way of defining gender roles determined by society and performed in accordance with the normative standards dictated by said society.

A fictitious diary about being in the process of painting, questioning doubting and exploring a gaze.

京都大学 / 0048 / 新制・課程博士 / 博士(情報学) / 甲第17967号 / 情博第511号 / 新制||情||91(附属図書館) / 30797 / 京都大学大学院情報学研究科知能情報学専攻 / (主査)教授 松山 隆司, 教授 乾 敏郎, 教授 石井 信 / 学位規則第4条第1項該当 / Doctor of Informatics / Kyoto University / DFAM

京都大学 / 0048 / 新制・課程博士 / 博士(人間健康科学) / 甲第18203号 / 人健博第20号 / 新制||人健||2(附属図書館) / 31061 / 京都大学大学院医学研究科人間健康科学系専攻 / (主査)教授 三谷 章, 教授 精山 明敏, 教授 髙橋 良輔 / 学位規則第4条第1項該当 / Doctor of Human Health Sciences / Kyoto University / DFAM

No description available.

In the recent years, a huge success has been accomplished in prediction of human eye fixations. Several studies employed deep learning to achieve high accuracy of prediction of human eye fixations. These studies rely on pre-trained deep learning for object classification. They exploit deep learning either as a transfer-learning problem, or the weights of the pre-trained network as the initialization to learn a saliency model. The utilization of such pre-trained neural networks is due to the relatively small datasets of human fixations available to train a deep learning model. Another relatively less prioritized problem is amount of computation of such deep learning models requires expensive hardware. In this dissertation, two approaches are proposed to tackle abovementioned problems. The first approach, codenamed DeepFeat, incorporates the deep features of convolutional neural networks pre-trained for object and scene classifications. This approach is the first approach that uses deep features without further learning. Performance of the DeepFeat model is extensively evaluated over a variety of datasets using a variety of implementations. The second approach is a deep learning saliency model, codenamed ClassNet. Two main differences separate the ClassNet from other deep learning saliency models. The ClassNet model is the only deep learning saliency model that learns its weights from scratch. In addition, the ClassNet saliency model treats prediction of human fixation as a classification problem, while other deep learning saliency models treat the human fixation prediction as a regression problem or as a classification of a regression problem.

No description available.