Showing papers on "Salience (neuroscience) published in 1998"

A model of saliency-based visual attention for rapid scene analysis

Abstract: A visual attention system, inspired by the behavior and the neuronal architecture of the early primate visual system, is presented. Multiscale image features are combined into a single topographical saliency map. A dynamical neural network then selects attended locations in order of decreasing saliency. The system breaks down the complex problem of scene understanding by rapidly selecting, in a computationally efficient manner, conspicuous locations to be analyzed in detail.

The representation of visual salience in monkey parietal cortex

Abstract: When natural scenes are viewed, a multitude of objects that are stable in their environments are brought in and out of view by eye movements. The posterior parietal cortex is crucial for the analysis of space, visual attention and movement 1 . Neurons in one of its subdivisions, the lateral intraparietal area (LIP), have visual responses to stimuli appearing abruptly at particular retinal locations (their receptive fields)2. We have tested the responses of LIP neurons to stimuli that entered their receptive field by saccades. Neurons had little or no response to stimuli brought into their receptive field by saccades, unless the stimuli were behaviourally significant. We established behavioural significance in two ways: either by making a stable stimulus task-relevant, or by taking advantage of the attentional attraction of an abruptly appearing stimulus. Our results show that under ordinary circumstances the entire visual world is only weakly represented in LIP. The visual representation in LIP is sparse, with only the most salient or behaviourally relevant objects being strongly represented.

The representation of visual salience in monkey parietal cortex

Abstract: When natural scenes are viewed, a multitude of objects that are stable in their environments are brought in and out of view by eye movements. The posterior parietal cortex is crucial for the analysis of space, visual attention and movement. Neurons in one of its subdivisions, the lateral intraparietal area (LIP), have visual responses to stimuli appearing abruptly at particular retinal locations (their receptive fields). We have tested the responses of LIP neurons to stimuli that entered their receptive field by saccades. Neurons had little or no response to stimuli brought into their receptive field by saccades, unless the stimuli were behaviourally significant. We established behavioural significance in two ways: either by making a stable stimulus task-relevant, or by taking advantage of the attentional attraction of an abruptly appearing stimulus. Our results show that under ordinary circumstances the entire visual world is only weakly represented in LIP. The visual representation in LIP is sparse, with only the most salient or behaviourally relevant objects being strongly represented.

Machine Learning of Generic and User-Focused Summarization

Abstract: A key problem in text summarization is finding a salience function which determines what information in the source should be included in the summary. This paper describes the use of machine learning on a training corpus of documents and their abstracts to discover salience functions which describe what combination of features is optimal for a given summarization task. The method addresses both "generic" and user-focused summaries.

Machine learning of generic and user-focused summarization

Abstract: A key problem in text summarization is finding a salience function which determines what information in the source should be included in the summary. This paper describes the use of machine learning on a training corpus of documents and their abstracts to discover salience functions which describe what combination of features is optimal for a given summarization task. The method addresses both "generic" and user-focused summaries.

A measure of motion salience for surveillance applications

Abstract: A measure of motion salience is proposed for surveillance applications. In this context, salient motion is taken as motion that is likely to result from a typical surveillance target (e.g., a person or vehicle travelling with a sense of direction through a scene) as opposed to other distracting motions (e.g., the scintillation of specularities on water, the oscillation of trees in the wind). The measure that is proposed makes use of spatiotemporal and gradient filters to characterize the extent to which a single coherent motion dominates local regions in space-time. Empirical results show that the measure is capable of making the desired distinctions. The significance of the approach lies in its ability to provide information about motion salience based on relatively simple, early vision operations. Such information can provide vital input to subsequent processing in attempting to distinguish targets of interest from mere distractors.

On the relative salience of Euclidean, affine, and topological structure for 3-D form discrimination.

Abstract: A match-to-sample task was performed, in which observers compared configurations of line segments presented stereoscopically in different three-dimensional orientations. Several different structural properties of these configurations were manipulated, including the relative orientations of line segments (a Euclidean property), their coplanarity (an affine property), and their patterns of cointersection (a topological property). Although the differences in these properties to be detected were all metrically equivalent, they varied dramatically in their relative perceptual salience, such that the error rates and reaction times in the three conditions varied by as much as 400%. Performance was highest in the topological condition, intermediate in the affine condition, and lowest in the Euclidean condition. These findings suggest that the relative perceptual salience of object properties may be systematically related to their structural stability under change, in a manner that is similar to the Klein hierarchy of geometries.

Connectedness and part-relation integration in shape category learning

Abstract: We investigated the role of connectedness in the use of part-relation conjunctions for object category learning. Participants learned categories of two-part objects defined by the shape of one part and its location relative to the other (part-relation conjunctions). The topological relationship between the parts (connected, separated, or embedded) varied between participants but was invariant for any given participant. In Experiment 1, category learning was faster and more accurate when an object’s parts were connected than when they were either separated or embedded. Subsequent experiments showed that this effect is not due to conscious strategies, differences in the salience of the individual attributes, or differences in the integrality/separability of dimensions across stimuli. The results suggest that connectedness affects the integration of parts with their relations in object category learning.

Automatic generation of subway directions: salience gradation as a factor fordetermining message and form

Abstract: A frequently encountered problem in urban life is navigation. In order to get to some place we use private means or public transportation, and if we lack clear directions we tend to ask for help. We will deal in this paper with the descriptions of subway routes and their automatic generation. In particular, we will try to show how the relative importance of a given piece of information can effect not only the message but also the form. 1 I n t r o d u c t i o n : the prob lem A frequently encountered •problem in urban life is the use of •public transportation: we have to get from here to there and quite so often we dori't know how. As it is not always possible to get help from a person (be it because nobody is available, or able to speak our language), we might appreciate assistance of a machine. In order to convey to the user "useful information", we must define what "usefulness" means. For example, if we •tell someone how to get from one place to another, we hardly ever specify all the intermediate steps, in particular if there is no ambiguity. Also, not all information is of equal weight. Yet, as we •will show ' the notion of "relative importance" of information is gradual in nature • rather than discrete , that is a simple binary Value (important vs. unimportant) 1. • All this reflects, of course, in the content and form of the final text. Relative importance is signaled by different means at the text level (headers, paragraphs, etc.) and at the sentence level (word choice, •syntactic structure: main clause versus subordinate clause, topic-comment •structures). • Concerning the prominence status (i.e. relative importance of a piece of information), semioticians and text linguists have reached a similar conclusion by distinguishing between the 'Yoreground/background" or "primary/secondary level" of a text [Bar66, vD77, AP89, Com92]. According to Combettes [Com92], the "primary level" deals with the core meaning, i.e. events and facts that make the text progress, while th e "secondary level" deaIs •with descriptions, evaluations, comments, and:reformulati0ns. " . ~: " i ~. : : : The distinction of levels, with information Of varying shades (salience gradation), implies t h a t it should be possible to identify corresponding linguistic "markers" for each one of them. Yet, as Combettes has pointed out [Com92], the means used for marking the relative importance of information may vary •from one type of text to another. Nevertheless, certain markers do •hold lln this respect we cieviate from:most,. current generation systems.

No role for motion blur in either motion detection or motion-based image segmentation

Abstract: The human contrast sensitivity function is bandpass in form for stimuli of low temporal frequency but low pass for flickering or moving stimuli. Because the loss in sensitivity to moving stimuli is large, images moving on the retina have little perceptible high-spatial-frequency content. The loss of high-spatial-frequency content—often referred to as motion blur—provides a potential cue to motion. The amount of motion blur is a function of stimulus velocity but is significant at velocities encountered by the visual system in everyday situations. Our experiments determined the influence of high-spatial-frequency losses induced by motion of this order on motion detection and on motion-based image segmentation. Motion detection and motion-based segmentation tasks were performed with either spectrally low-pass or spectrally broadband stimuli. Performance on these tasks was compared with a condition having no motion but in which form differences mimicked the perceptual loss of high spatial frequencies produced by motion. This allowed the relative salience of motion and motion-induced blur to be determined. Neither image segmentation nor motion detection was sensitive to the high-spatial-frequency content of the stimuli. Thus the change in perceptual form produced in moving stimuli is not normally used as a cue either for motion detection or for motion-based image segmentation in ordinary situations.

Cortical synchronization and perceptual salience

Abstract: We present a striate-cortical model which proposes a direct relationship between cellular synchronization and perceptual salience. The model focuses on the role of the longrange horizontal connections between oriented simple cells in striate cortex and is able to account for current physiological and psychophysical results on contour salience. We demonstrate that horizontal connections between realistically-modeled multi-compartment pyramidal cells and interneurons can generate robust context-dependent synchronization. Closed contours induce better synchronization in the network than open contours, and closure thus increases perceptual salience, as observed psychophysically by Kovacs and Julesz. This result is a general topological property of synchronization. The model supports a temporal synchronization solution to the binding problem, in that changes in synchronization are directly linked to changes in visual perception.

Modeling cognitive effects on visual search for targets in cluttered backgrounds

Abstract: To understand how a human operator performs visual search in complex scenes, it is necessary to take into account top- down cognitive biases in addition to bottom-up visual saliency effects. We constructed a model to elucidate the relationship between saliency and cognitive effects in the domain of visual search for distant targets in photo- realistic images of cluttered scenes. In this domain, detecting targets is difficult and requires high visual acuity. Sufficient acuity is only available near the fixation point, i.e. in the fovea. Hence, the choice of fixation points is the most important determinant of whether targets get detected. We developed a model that predicts the 2D distribution of fixation probabilities directly from an image. Fixation probabilities were computed as a function of local contrast (saliency effect) and proximity to the horizon (cognitive effect: distant targets are more likely to be found c close to the horizon). For validation, the model's predictions were compared to ensemble statistics of subjects' actual fixation locations, collected with an eye- tracker. The model's predictions correlated well with the observed data. Disabling the horizon-proximity functionality of the model significantly degraded prediction accuracy, demonstrating that cognitive effects must be accounted for when modeling visual search.

The meeting place problem: Salience and search.

Abstract: The notion of salience was developed by Schelling in the context of the meeting-place problem of locating a partner in the absence of a pre-agreed meeting place. In this paper, we argue that a realistic specification of the meeting place problem involves allowing a strategy of active search over a range of possible meeting places. We solve this extended problem, allowing for extensions such as repeated play, search costs and asymmetric payoffs. The result is a considerably richer, but more complex, notion of salience. (C) 1998 Elsevier Science B.V.