Citation Context

...aliency maps. In order to obtain an objective comparison of segmentation results, we use a ground truth image database. We derived the database from the publicly available database used by Liu et al. =-=[20]-=-. This database provides bounding boxes drawn around salient regions by nine users. However, a bounding box-based ground truth is far from accurate, as also stated by Wang and Li [28]. Thus, we create...

Citation Context

...are then extended using a fuzzy growth model. Harel et al. [12] normalize Itti and colleagues’ feature maps to highlight conspicuous parts and admit combination with other importance maps. Liu et al. =-=[18]-=- find multi-scale contrast by linearly combining contrast in a Gaussian image pyramid. More recently, Goferman et al. [10] simultaneously model local low-level clues, global considerations, visual org...

Citation Context

...small parts of objects. We argue that any object has at least one of three distinctive characteristics: (a) a well-defined closed boundary in space; (b) a different appearance from their surroundings =-=[23, 25]-=-; (c) sometimes it is unique within the image and stands out as salient [2, 13, 17, 19]. Many objects have several of these characteristics at the same time (fig. 2-4). This paper makes three contribu...

Citation Context

... type of saliency is important for understanding human attention as well as for specific applications such as auto focusing. Others have concentrated on detecting a single dominant object of an image =-=[13, 7, 5]-=-. For instance, in Figure 1, such 1 These descriptions were obtained by collecting titles given by 12 different people. See samples in the second row of Figure 1 Figure 1. Our context-aware saliency r...

Citation Context

...nabling the system to localize many types of objects. See section 3 for a more detailed overview. An alternative approach is to attempt to segment pixels of foreground objects [22] or salient regions =-=[23, 24]-=-. However, these approaches may not be suitable for localizing individual objects in cluttered scenes, because a continuous foreground or salient region may contain many objects. Two concurrent works ...

Citation Context

...meets the real-time requirements, which has great potentials in engineering applications. 1. Introduction Most traditional object detectors need training in order to detect specific object categories =-=[1, 2, 3]-=-, but human vision can focus on general salient objects rapidly in a clustered visual scene without training because of the existence of visual attention mechanism. So human can easily deal with gener...

Citation Context

...on a combination of insights from neurosciences, biology, computer vision, and other fields. However, recent research has shown that computational models simulating low-level stimuli-driven attention =-=[17, 20, 21]-=- are quite successful and represent useful tools in many application scenarios, including image segmentation [14], resizing [5] and object detection [27]. Results from perceptual research [11,24,25] i...

Citation Context

...discovery. Whereas most saliency detectors rely on bottom-up image cues (e.g., [12, 9]), recent work suggests that higher-level saliency may actually be learned from labeled data of segmented objects =-=[19, 1, 6, 8]-=-, drawing on classic Gestalt cues. In particular, interesting approaches to generate and rank an image’s multiple figure-ground segmentation hypotheses are explored in [6, 8], with results showing tha...

Citation Context

...tiple important objects. Saliency detection Early saliency detectors rely on bottom-up image cues (e.g., [12]). More recent work tries to learn high-level saliency measures, whether for static images =-=[18, 3, 6]-=- or video [16]. Whereas typically such metrics aim to prime a visual search process, we are interested in high-level saliency for the sake of isolating those things worth summarizing. Researchers have...