Suppr超能文献

无特征梯度下的视觉显著性与纹理分离

Visual saliency and texture segregation without feature gradient.

作者信息

Ben-Shahar Ohad

机构信息

Department of Computer Science and the Zlotowski Center for Neuroscience, Ben-Gurion University, Beer Sheva 84105, Israel.

出版信息

Proc Natl Acad Sci U S A. 2006 Oct 17;103(42):15704-9. doi: 10.1073/pnas.0604410103. Epub 2006 Oct 9.

Abstract

A central notion in the study of texture segregation is that of feature gradient (or feature contrast). In orientation-based texture segregation, orientation gradients have indeed played a fundamental role in explaining behavioral results. Here, however, we show that general, smoothly varying, orientation-defined textures (ODTs) exhibit striking perceptual singularities that are completely unpredictable from orientation gradients. These singularities defy not only popular texture segregation theories but also virtually all computational segmentation methods, and they confound previous behavioral studies with smoothly varying ODTs. We provide psychophysical evidence that perceptual singularities in smooth ODTs are salient visual features consistent across observers and with significant effect on the perception and segregation of oriented textures. We further show that, although orientation gradients cannot predict them, perceptual singularities in smooth ODTs emerge directly from, and can be spatially localized by, two ODT curvatures. Given the traditional role of feature gradients in early vision, the significance of these findings extends well beyond orientation-based texture segregation to issues ranging from curve integration and fragment grouping, through the perception of 3D shape, to the functional organization of the primary visual cortex.

摘要

纹理分离研究中的一个核心概念是特征梯度(或特征对比度)。在基于方向的纹理分离中,方向梯度确实在解释行为结果方面发挥了重要作用。然而,在这里我们表明,一般的、平滑变化的、基于方向定义的纹理(ODT)呈现出惊人的感知奇点,这些奇点从方向梯度来看是完全不可预测的。这些奇点不仅违背了流行的纹理分离理论,而且几乎违背了所有的计算分割方法,并且它们使先前关于平滑变化的ODT的行为研究变得混乱。我们提供了心理物理学证据,表明平滑ODT中的感知奇点是观察者之间一致的显著视觉特征,并且对定向纹理的感知和分离有显著影响。我们进一步表明,虽然方向梯度无法预测它们,但平滑ODT中的感知奇点直接源于两个ODT曲率,并且可以通过这两个曲率在空间上定位。鉴于特征梯度在早期视觉中的传统作用,这些发现的意义远远超出了基于方向的纹理分离,涵盖了从曲线整合和片段分组、3D形状感知到初级视觉皮层功能组织等一系列问题。

相似文献

1
Visual saliency and texture segregation without feature gradient.无特征梯度下的视觉显著性与纹理分离
Proc Natl Acad Sci U S A. 2006 Oct 17;103(42):15704-9. doi: 10.1073/pnas.0604410103. Epub 2006 Oct 9.
9
Discrimination of orientation-defined texture edges.方向定义纹理边缘的辨别
Vision Res. 1995 Oct;35(20):2863-77. doi: 10.1016/0042-6989(94)00302-3.

引用本文的文献

4
Curvature processing dynamics in macaque area V4.猴 V4 区的曲率加工动力学。
Cereb Cortex. 2013 Jan;23(1):198-209. doi: 10.1093/cercor/bhs004. Epub 2012 Jan 31.
5
Estimation of 3D shape from image orientations.从图像方向估计 3D 形状。
Proc Natl Acad Sci U S A. 2011 Dec 20;108(51):20438-43. doi: 10.1073/pnas.1114619109. Epub 2011 Dec 6.
6
Analogous intermediate shape coding in vision and touch.视觉与触觉中类似的中间形状编码。
Proc Natl Acad Sci U S A. 2009 Sep 22;106(38):16457-62. doi: 10.1073/pnas.0904186106. Epub 2009 Sep 4.

本文引用的文献

1
On the surprising salience of curvature in grouping by proximity.关于在按接近度分组中曲率令人惊讶的显著性。
J Exp Psychol Hum Percept Perform. 2006 Apr;32(2):226-234. doi: 10.1037/0096-1523.32.2.226.
2
Hue geometry and horizontal connections.色调几何与水平连接。
Neural Netw. 2004 Jun-Jul;17(5-6):753-71. doi: 10.1016/j.neunet.2004.03.011.
5
Population coding of shape in area V4.V4区中形状的群体编码
Nat Neurosci. 2002 Dec;5(12):1332-8. doi: 10.1038/nn972.

文献AI研究员

20分钟写一篇综述,助力文献阅读效率提升50倍。

立即体验

用中文搜PubMed

大模型驱动的PubMed中文搜索引擎

马上搜索

文档翻译

学术文献翻译模型,支持多种主流文档格式。

立即体验