Interesting objects are visually salient.

By Lior Elazary, Laurent Itti. Journal of Vision, 2008

305 citations recorded by the source corpus.

How do we decide which objects in a visual scene are more interesting? While intuition may point toward high-level object recognition and cognitive processes, here we investigate the contributions of a much simpler process, low-level visual saliency. We used the LabelMe database (24,863 photographs with 74,454 manually outlined objects) to evaluate how often interesting objects were among the few most salient locations predicted by a computational model of bottom-up attention. In 43% of all images the model's predicted most salient location falls within a labeled region (chance 21%). Furthermore, in 76% of the images (chance 43%), one or more of the top three salient locations fell on an outlined object, with performance leveling off after six predicted locations. The bottom-up attention model has neither notion of object nor notion of semantic relevance. Hence, our results indicate that selecting interesting objects in a scene is largely constrained by low-level visual properties rather than solely determined by higher cognitive processes.

Subfields: Cognitive Neuroscience, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/611c084900a847781f77e8b563113256b060d9f1

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Cognitive Neuroscience

In Computer Vision and Pattern Recognition

Citations in the corpus

Bringing Semantics into Focus Using Visual Abstraction (2013, methodology): “Many methods use tagged objects in images to predict important objects directly from visual features [1, 5, 16, 31], and to study the properties of popular tags [1, 31].”
Learning the Relative Importance of Objects from Tagged Images for Retrieval and Cross-Modal Search (2012, background): “Though difficult to state in the absolute, a number of previous studies lend support for both points (von Ahn and Dabbish 2004; Spain and Perona 2008; Einhauser et al. 2008; Elazary and Itti 2008; Tatler et al. 2005; Hwang and Grauman 2010b), as does our own experimental data.”
Salient object detection: A survey (2014, background): “ds, an object is more likely to be salient than a region on the background. This is based on a ﬁnding that image background is usually more structured and homogeneous (thus less salient) than objects [42], [45]. On the other hand, the latter approaches use objectness measures to assign higher salien”
What do eyes reveal about the mind?: Algorithmic inference of search targets from fixations (2015, background): “, [27,14]), we qualitatively observe that viewers look more often at the”
Modeling Search for People in 900 Scenes: A combined source model of eye guidance (2009, unknown): “In natural images, it is interesting to note that objects are typically more salient than their background ( Elazary & Itti, 2008; Torralba, Oliva, Castelhano, & Henderson, 2006), so oculomotor guidance processes may use saliency as a heuristic to fixate objects in the scene rather than the backgro”
We get the algorithms of our ground truths: Designing referential databases in digital image processing (2017, unknown): “See Elazary and Itti (2008); Itti and Koch (2001); Itti et al. (1998, 2000); Zhao and Koch (2011).”
Understanding and predicting importance in images (2012, background): “Predicting Importance: Elazary and Itti [4] consider object naming order in the LabelMe dataset [21] as a measure of the interestingness of an object and compare that to salient locations predicted by computational models of bottom-up attention.”
Computational visual attention systems and their cognitive foundations: A survey (2010, methodology): “On the other hand, there are cases in which an eye movement is not preceeded by covert attention: Findlay and Gilchrist [2001] found that in tasks like reading and complex object search, saccades (quick, simultaneous movements of both eyes in the same direction [Cassin and Solomon 1990]) were made w”
Salient Object Detection: A Discriminative Regional Feature Integration Approach (2013, methodology): “The analysis (Elazary and Itti 2008) over the LabelMe annotation data (Russell et al.”
Some Objects Are More Equal Than Others: Measuring and Predicting Importance (2008, background): “Elazary and Itti consider the object annotation order a measure of interestingness [14]; however, as partial annotations are passed on to new users, a single ordering is produced.”
Adopting Abstract Images for Semantic Scene Understanding (2016, background): “They allow for the direct study of how to infer high-level semantic information, since they remove the reliance on noisy low-level object, attribute and relation detectors, or the tedious hand-labeling of real images.”
Attentive Systems: A Survey (2017, background): “As discussed in Elazary and Itti (2008), interesting objects are visually salient.”
Predicting eye fixations using convolutional neural networks (2015, background): “The proposed approach achieves state-of-the-art results over four publically available benchmark datasets, demonstrating the superiority of our work.”
Creativity, attention and the memory for brands: an outdoor advertising field study (2015, methodology): “An earlier study by Elazary and Itti (2005) used a similar but different database of images to show that objects identified as interesting correlate strongly with the saliency map.”
VIP: Finding important people in images (2015, background): “This reasoning is across a corpus of photos (each containing a person of interest), and the objective is to assign an importance score to each image.”
Visual Saliency Based on Scale-Space Analysis in the Frequency Domain (2013, background): “However, saliency detection algorithms in computer vision are assumed and expected to have the ability to detect salient object regions in a scene [45].”
A Benchmark of Computational Models of Saliency to Predict Human Fixations (2012, background): “Several authors have shown that regions of interest found by computational model of visual attention correlate with eye fixations [59] [53] [19] [29] [52] [6] [33] [57] and the reported area under receiver operating characteristic (ROC) curve, which measures how well the two correlate, has increased”
Region-Based Saliency Detection and Its Application in Object Recognition (2014, methodology): “It is also validated by some neurobiological finding that objects predict fixations better than early saliency [17], [18].”
Exploiting local and global patch rarities for saliency detection (2012, background): “Also it has been shown that interesting objects are more salient within a scene, providing support in favor of object-based attention [43].”