In this paper we present a technique for saliency estimation in 360 images. Existing approaches exploit high/low-level image features, head movement, and eye-gazes of observers. However, we believe that the saliency of an image is influenced also by its content. The proposed approach consists of three modules. In the first module, a segmentation of the image is performed and its local and global features are extracted. The second module combines the extracted features for summarizing the content that the image portrays. Finally, this content knowledge is utilised for image saliency estimation. Experimental results show effectiveness of the proposed system with respect to ground truth saliency maps.
Mazumdar, P., Battisti, F. (2019). A Content-Based Approach for Saliency Estimation in 360 Images. In Proceedings - International Conference on Image Processing, ICIP (pp.3197-3201). IEEE Computer Society [10.1109/ICIP.2019.8803296].