Improving Visual Saliency by Adding ‘Face Feature Map’ and ‘Center Bias’

Marat, Sophie; Rahman, Anis Ur; Pellerin, Denis; Guyader, Nathalie; Houzet, Dominique

doi:10.1007/s12559-012-9146-3

Cited by 54 publications

(43 citation statements)

References 45 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This bias is currently introduced in computational modeling through ad hoc methods. In (Le Meur et al, 2006;Marat et al, 2013), the saliency map is simply multiplied by a 2D anisotropic…”

Section: Discussionmentioning

confidence: 99%

Introducing context-dependent and spatially-variant viewing biases in saccadic models

Meur

Coutrot

2016

Vision Research

View full text Add to dashboard Cite

Previous research showed the existence of systematic tendencies in viewing behavior during scene exploration. For instance, saccades are known to follow a positively skewed, long-tailed distribution, and to be more frequently initiated in the horizontal or vertical directions. In this study, we hypothesize that these viewing biases are not universal, but are modulated by the semantic visual category of the stimulus. We show that the joint distribution of saccade amplitudes and orientations significantly varies from one visual category to another. These joint distributions are in addition spatially variant within the scene frame. We demonstrate that a saliency model based on this better understanding of viewing behavioral biases and blind to any visual information outperforms well-established saliency models. We also propose a saccadic model that takes into account classical low-level features and spatially-variant and context-dependent viewing biases. This model outperforms state-of-the-art saliency models, and provides scanpaths in close agreement with human behavior. The better description of viewing biases will not only improve current models of visual attention but could also influence many other applications such as the design of human-computer interfaces, patient diagnosis or image/video processing applications.

show abstract

“…This bias is currently introduced in computational modeling through ad hoc methods. In (Le Meur et al, 2006;Marat et al, 2013), the saliency map is simply multiplied by a 2D anisotropic…”

Section: Discussionmentioning

confidence: 99%

Introducing context-dependent and spatially-variant viewing biases in saccadic models

Meur

Coutrot

2016

Vision Research

View full text Add to dashboard Cite

show abstract

“…Nevertheless, the model does not include any static cues as colors, for example. Description: The STVSM model [10] is inspired by the biology of the visual system and breaks down each frame of a video into three maps: a static saliency map emphasizes regions that differ from their context in terms of luminance, orientation, and spatial frequency. A dynamic saliency map emphasizes moving regions with values proportional to motion amplitude.…”

Section: Smams: Saliency Models For Abnormal Motion Selection (2011)mentioning

confidence: 99%

“…GBVS [ 3 ] NMPT [ 4 ] SSOV [ 5 ] SDSR [6 ] VICO [ 7 ] SMVQA [ 8 ] SMAMS [9] STVSM [ 10 ] Fig. 10.1 Chronological overview of salient models for videos Indeed, camera motions has a great impact on saliency estimation, and models need to be specifically designed to manage the temporal aspect.…”

mentioning

confidence: 99%

Bottom-Up Saliency Models for Videos: A Practical Review

Riche

Mancaş

2016

From Human Attention to Computational Attention

View full text Add to dashboard Cite

Research on visual saliency initially focused on still images rather than on video content. However, in the recent years, an increasing demand of video saliency appeared for some applications like gaming, editing, video retargeting, smart TV, robot navigation, surveillance, etc. Therefore, remarkable progress has been made first in the understanding on eye tracking data with dynamical stimuli and, in a second time, in the modeling process.There are fundamental differences between videos and still images. For example, each video frame is only observed during a fraction of a second, while a still image can be viewed much longer. Some videos can feature varying camera motion such as tilting, panning, zooming, etc. For this reason, videos are probably viewed differently by human observers than still images, and some comprehensive comparative studies have emerged. In [1], for example, the authors study the influence of tasks on gaze behavior in static and dynamic scenes. In [2], the gaze on static and dynamic scene is compared; it also shows that the center bias decreases with dynamic stimuli.In terms of modeling, static models have first been extended to video. This is the case for GBVS, SDSR, NMPT, or SSOI where authors added dynamic features to their models. Though these existing models are major contributions, video saliency estimation methods should then differ substantially from image saliency methods.

show abstract

“…In a similar vein, object knowledge can be used to top-down tune early salience. In particular, when dealing with faces within the scene, a face detection step can provide a reliable cue to complement early conspicuity maps, as it has been shown by Cerf et al [21], deCroon et al [27], Marat et al [65], or a useful prior for Bayesian integration with low level cues [13]. This is indeed an important issue since faces may drive attention in a direct fashion [20].…”

Section: Levels Of Representation and Controlmentioning

confidence: 99%

“…In this paper we are not much involved in discussing neurobiological underpinnings of computational theories, but, interestingly enough the approach of fusing object-based information with low-level salience, either through straightforward combination [21,65] or in the formal framework of Bayesian modelling [13,24,15] provides a computational account of the way the lateral intraparietal area (LIP) of posterior parietal cortex acts as a priority map to guide the allocation of covert attention and eye movements (overt attention). The LIP is a cortical area located at the interface between visual input and oculomotor output and it is well known that LIP activity is biased by both bottom-up stimulus driven factors and top-down cognitive influences.…”

Section: Levels Of Representation and Controlmentioning

confidence: 99%

Modelling Task-Dependent Eye Guidance to Objects in Pictures

et al. 2014

View full text Add to dashboard Cite

We introduce a model of attentional eye guidance based on the rationale that the deployment of gaze is to be considered in the context of a general actionperception loop relying on two strictly intertwined processes: sensory processing, depending on current gaze position, identifies sources of information that are most valuable under the given task; motor processing links such information with the oculomotor act by sampling the next gaze position and thus performing the gaze shift. In such a framework, the choice of where to look next is task-dependent and oriented to classes of objects embedded within pictures of complex scenes. The dependence on task is taken into account by exploiting the value and the payoff of gazing at certain image patches or proto-objects that provide a sparse representation of the scene objects. The different levels of the actionperception loop are represented in probabilistic form and eventually give rise to a stochastic process that generates the gaze sequence. This way the model also accounts for statistical properties of gaze shifts such as individual scan path variability. Results of the simulations are compared either with experimental data

show abstract

Improving Visual Saliency by Adding ‘Face Feature Map’ and ‘Center Bias’

Cited by 54 publications

References 45 publications

Introducing context-dependent and spatially-variant viewing biases in saccadic models

Introducing context-dependent and spatially-variant viewing biases in saccadic models

Bottom-Up Saliency Models for Videos: A Practical Review

Modelling Task-Dependent Eye Guidance to Objects in Pictures

Contact Info

Product

Resources

About