Statistical efficiency approach to visual representation and object recognition

视觉表示和对象识别的统计效率方法

• 批准号: 09610072
• 负责人: ISHIGUCHI Akira
• 金额: $ 2.05万
• 依托单位: Ochanomizu University
• 依托单位国家: 日本
• 项目类别: Grant-in-Aid for Scientific Research (C)
• 财政年份: 1997
• 资助国家: 日本
• 起止时间: 1997 至 1998
• 项目状态: 已结题
• 来源: https://kaken.nii.ac.jp/grant/KAKENHI-PROJECT-09610072/
• 关键词: object recognition; ideal observer; efficiency; signal detection theory; symmetry; recovery of structure; stereo vision; motion; 視点; 理想観察者; ガウスノイズ

This study investigated the ability of human visual system to recognize a 3D object and its representation in the brain by use of a statistical efficiency method. In principle, we added some spatial Gaussian noise to stimuli, and calculated the statistical efficiency with comparing the discriminability of human observer with that of the ideal observer. The results of this research are as follows :(1) We presented wire-frame models of some 3D objects in a computer display and made subjects observe them with the monocular or the stereoscopic view. The results showed that a structure of the object had effects in the discrimination of the object under the stereoscopic vision.(2) We investigated the ''recovery of structure from motion" with use of the efficiency method. The results showed that the recovery of 3D structure from motion was less than that of 2D structure, and suggested that sampling errors occurr in a process of the 3D recovery.(3) We investigated an ability of the visual system to recognize a 3D object overlapped some spatial noises under the stereoscopic condition. It was suggested that, in this situation, a process of grouping of the components could be involved in the object recognition.(4) We investigated the segregation of moving objects from background moving noises with use of the efficiency method. The difference of direction or velocity of the moving object from the background noises was crucial factors in the segregation. The results showed low efficiencies of the visual system in the discrimination of the moving object. This indicated that the visual system samples the motion information with low precision.

本研究采用统计效率方法研究了人类视觉系统对三维物体的识别能力及其在大脑中的表征。原则上，我们在刺激中加入空间高斯噪声，并通过比较人类观察者和理想观察者的分辨能力来计算统计效率。本研究的主要结果如下：（1）在计算机显示器上显示了一些三维物体的线框模型，并使被试者从单目或立体的角度观察这些物体。结果表明，在立体视觉条件下，物体的结构对物体的辨别有影响。(2)我们用效率方法研究了“从运动恢复结构”问题。结果表明，从运动图像中恢复三维结构的速度要比恢复二维结构的速度慢，这说明在三维结构的恢复过程中存在采样误差。(3)本文研究了在立体视觉条件下，视觉系统对叠加有空间噪声的三维物体的识别能力。据建议，在这种情况下，物体识别可涉及对各组成部分进行分组的过程。(4)我们研究了分离的运动目标从背景运动噪声与使用的效率方法。运动目标的运动方向或运动速度与背景噪声的差异是造成分离的关键因素。结果表明，视觉系统对运动物体的识别效率较低。这表明视觉系统以低精度对运动信息进行采样。