Psychophysical and computational study on feature integration and stimulus context dependence in the perception of object attributes.

我们研究了视觉皮层如何整合提取的特征来感知物体属性。在本研究中，我们提出了一个计算模型的皮质架构，集成的功能，基于心理物理学的研究结果和详细的生理知识。具体来说，我们提出了一个假设，即皮层过程的每个阶段提取的特征是基于该阶段的上下文相关性进行整合的，计算模型包括V1和V2区域的上下文相关性，包括（1）线方向的提取，（2）通过周围抑制和易化实现的上下文相关性，（3）边界所有权方向的确定。我们定量地研究了这个模型是否再现了轮廓和表面的感知，与生理和心理物理的发现非常相似，包括V1神经元的上下文调制的模型显示了与一些倾斜错觉一致的模拟结果。V2模型表明，从V1神经元提取的对比度信息能够确定图形的方向，而无需检测T连接和角点。该模型表现出一致和强大的反应，是明显的轮廓和方向的图形的感知，并再现生理和心理物理学的主要特征。

We investigated how the visual cortex integrates extracted features to perceive the object attribute. In the present study, we proposed a computational model of the cortical architecture that integrated the features, based on psychophysical findings and detailed physiological knowledge. Specifically, we proposed the hypothesis that the extracted features within each stage of cortical processes are integrated based on the contextual dependency of the stage.The computational model includes the contextual dependency of V1 and V2 regions, which consists of (1)extraction of line orientation, (2)contextual dependency realized by surrounding suppression and facilitation, and (3)determination of the direction of border ownership. We examined quantitatively whether this model reproduced the perception of contours and surfaces with close resemblance to physiological and psychophysical findings.The model that include contextual modulation of V1 neurons showed the simulation results consistent with a number of tilt illusions. The model of V2 showed that contrast information extracted from V1 neurons were capable of determining the direction of figure without the detection of T-junctions and corners. The proposed model showed consistent and robust responses that are apparent in the perception of contour and the direction of figure, and reproduced major characteristics of physiology and psychophysics.