Dövencioğlu Dicle N, Welchman Andrew E, Schofield Andrew J
School of Psychology, University of Birmingham, Edgbaston, Birmingham, UK.
Vision Res. 2013 Jan 25;77:1-9. doi: 10.1016/j.visres.2012.11.005. Epub 2012 Nov 28.
Luminance variations are ambiguous: they can signal changes in surface reflectance or changes in illumination. Layer decomposition-the process of distinguishing between reflectance and illumination changes-is supported by a range of secondary cues including colour and texture. For an illuminated corrugated, textured surface the shading pattern comprises modulations of luminance (first order, LM) and local luminance amplitude (second-order, AM). The phase relationship between these two signals enables layer decomposition, predicts the perception of reflectance and illumination changes, and has been modelled based on early, fast, feed-forward visual processing (Schofield et al., 2010). However, while inexperienced viewers appreciate this scission at long presentation times, they cannot do so for short presentation durations (250 ms). This might suggest the action of slower, higher-level mechanisms. Here we consider how training attenuates this delay, and whether the resultant learning occurs at a perceptual level. We trained observers to discriminate the components of plaid stimuli that mixed in-phase and anti-phase LM/AM signals over a period of 5 days. After training, the strength of the AM signal needed to differentiate the plaid components fell dramatically, indicating learning. We tested for transfer of learning using stimuli with different spatial frequencies, in-plane orientations, and acutely angled plaids. We report that learning transfers only partially when the stimuli are changed, suggesting that benefits accrue from tuning specific mechanisms, rather than general interpretative processes. We suggest that the mechanisms which support layer decomposition using second-order cues are relatively early, and not inherently slow.
它们既可以表示表面反射率的变化，也可以表示光照的变化。层分解——区分反射率和光照变化的过程——得到了一系列包括颜色和纹理在内的次要线索的支持。对于一个被照亮的波纹状、有纹理的表面，阴影图案包括亮度调制（一阶，LM）和局部亮度幅度（二阶，AM）。这两个信号之间的相位关系能够实现层分解，预测对反射率和光照变化的感知，并且已经基于早期、快速、前馈视觉处理进行了建模（斯科菲尔德等人，2010年）。然而，虽然没有经验的观察者在长时间呈现时能够理解这种分离，但在短呈现持续时间（250毫秒）时却做不到。这可能表明存在更慢、更高层次的机制在起作用。在这里，我们考虑训练如何减少这种延迟，以及由此产生的学习是否发生在感知层面。我们训练观察者在5天的时间里区分混合了同相和反相LM/AM信号的格子图案刺激的成分。训练后，区分格子图案成分所需的AM信号强度大幅下降，表明发生了学习。我们使用具有不同空间频率、面内方向和锐角格子图案的刺激来测试学习的迁移。我们报告说，当刺激改变时，学习仅部分迁移，这表明益处来自于调整特定机制，而不是一般的解释过程。我们认为，使用二阶线索支持层分解的机制相对较早，并非本质上就慢。
