Serbest görüntüleme ve görev güdümlü görsel dikkat-çekerlik çalışması

Visual saliency is a widely studied field in computer science. The visual saliency is a field concerning people's visual attention while seeing visual stimuli, an image or frames in a video sequence, and visual saliency methods aim at estimating people's eye fixations correctly. While humans have a granted and comprehensive sight ability provided by the human visual system (HVS), it is not effortless for visual saliency approaches to catch eye fixation points easily and very accurately. Thus, there may be different visual saliency approaches to handle different types of visual attention, affected by viewing conditions or goals of the viewing. In this thesis, we determine to analyze proposed visual saliency methods for different viewing conditions and different attention types, i.e. bottom-up and top-down. Bottom-up visual attention emerges when people observe visual stimuli freely. Topdown visual attention appears when people view visual stimuli related to the content and the viewer has a goal to consider while viewing the visual stimuli. In the first part of the thesis, we explain our study in which we examine state-ofthe-art visual saliency methods for 2D desktop and 3D VR viewing conditions. In the second part of the thesis, we focus on top-down visual attention and integrate visual saliency predictions for different goals of the viewers into a single Generative Adversarial Network (GAN) [1] structure.