Saliency and Object Detection

Visual attention allows the human visual system to effectively deal with the huge flow of visual information acquired by the retina. Since the years 2000, the human visual system began to be modelled in computer vision and it became part of artificial intelligence: while learning focuses on repetitive data which can easily be modeled, computational attention focuses on abnormal, rare and surprising data which can hardly be learnt. Attention is a product of the continuous interaction between bottom-up and top-down information. While the bottom-up information has been extensively investigated through saliency models, top-down influence on visual attention has been less investigated. This paper intends to study the influence of object-based (faces and text) top-down information on bottom-up saliency maps. It proposes a simple yet effective fusion scheme that can be applied on any bottom-up saliency model depending on the object detector effectiveness and the object size. The evaluation results show that it is possible to highly improve classical bottom-up saliency models with the arrival of better object detectors. In the future, such attention models can become as effective as deep-learning based attention models while keeping them more generic and avoiding underestimating bottom-up features.