Visual Attention Analysis and User Guidance in Cinematic VR Film

Due to the character of 360° video, it is often a challenge for filmmakers to guide the attention of users to the region of interest. Visual effects as a type of user guidance is frequently applied to traditional film. Nevertheless, the influence of visual effects in 360° video has been rarely explored. For this reason, the purpose of this paper is to study how four different visual effects, respectively Desaturation, Context-based Darkening, Area Darkening, and Object to Follow, affect visual attention of users in 360° video. Therefore, we performed a subjective test as well as analyzed the saliency maps predicted with a convolutional neural network. In the subjective test, 15 participants were requested to watch four 360° videos, which were implemented with visual effects, while the position of their viewport was recorded. The results of this work are compared to earlier research on the same videos without visual effects. We show that Area Darkening has the best effect on guiding the visual attention, Context-based Darkening makes the best contribution on enhancing the saliency of the region of interest, while Desaturation has nearly no effect for user guidance and does not change the saliency of the videos. A Logo as Object to Follow create a new salient area, while the predicted saliency of areas apart from the Logo remains the same.