Personalized Visual Attention Modelling

Miroslav Čulík

Supervisor(s): Miroslav Laco

Slovak Technical University


Abstract: The vast majority of models predicting visual attention are not designed to take observer-specific information into account. However, some models that use these features could be useful in predicting personalized attention, what can provide space for categorizing observers based on their visual attention properties. We introduce two variants of personalized models built upon the generalized model of Convolutional Neural Network (CNN). Predictions of these two personalized variants were evaluated and compared with the predictions of the generalized model. Our observations indicate that better results were produced by the model predicting personalized saliency maps for a particular observer.
Keywords: Computer Vision, Image Processing
Full text:
Year: 2021