PhD. Depth Exam From attention to object proposals

The problem of finding locations where people look at first in images, known as saliency detection, spans decades of research from multiple disciplines including psychology, neuroscience, and computer vision. Because of the complexity of the problem it can hardly be considered as solved. Here, we give an overview of the methods for saliency detection starting from early biologically-plausible methods from neuroscience and ending with supervised methods using machine-learning techniques in computer vision. We divide saliency detection methods into three main categories: psychologically-inspired, segmentation-based, and data-driven. We review representative works in each category. Also, we discuss challenges such as evaluation criteria, repeatability and robustness of the methods, and list a number of possible directions for future work.