Context Proposals for video saliency segmentation
Aymen Azaza, Rahma Kalboussi, Mehrez Abdellaoui, Ali Douik · 2024
Generally, salient object segmentation is broken up into two main tasks:segmentation and saliency assignment. Lately, using object proposals in many computer vision applications showed significant impact on object detection, image segmentation and recently saliency detection in still images. Neverless, they have not been evaluated for video saliency segmentation yet. Besides, the impact of context features on video saliency has not been evaluated yet. Consequently, we explore in this paper the use of context proposals for salient object segmentation in videos. We prove that using context features issued from object proposals is a good saliency indicator and provide good video saliency segmentation results. The proposed method was evaluated on two challenging benchmark datasets and showed better performance that the state-of-the-art.