Combined visual attention model for video sequences

The paper presents a model of visual attention combined with eye tracking to drive content-based retrieval of image data in order to facilitate understanding and development of new adaptive eye guided representation of image sequences. The bottom-up component of the proposed visual attention model is based on the extended Itti-Koch saliency model incorporating conjunction search and temporal aspects of sequences of natural images. The top-down component is a gaze-prediction model designed to associate measured eye tracking locations and features extracted from images. This approach permits the detection and separation of attention-driven regions of interest and their processing with the highest accuracy, while the remaining part of the image (the background) is reproduced with lower quality.