Construction of high level vision system with effective visual field

本研究的目的是构建高效视觉信息处理的视觉模型，该模型具有人类视觉加工阶段的“扫视视图”和“搜索视图”。然后，以人的心理学概念--有效视野为指标，选取适当的加工阶段，对人的视觉功能中的有效视野进行了心理学实验，并分析了其特点。将有效视场模型应用于传统的视觉系统，构成了新的视觉系统。通过仿真实验，验证了该模型的有效性。然后，对不同图案进行了有效视野的心理实验。结果表明，视觉加工问题的复杂性与有效视野之间存在着权衡关系。同时也说明了人类在瞬间通过某种方法从外部性中量化信息。然后，我们尝试构建高级视觉系统，通过在系统中包含这样的功能，可以选择视觉方向，有利于三维物体的识别。因此，作为表示被摄体图像的信息量的指标的被摄体图像熵被用作指标来代替它。在心理学实验中，证实了熵值越大，越容易被识别。另外，通过计算机对三维物体进行建模，实现了视觉方向的自动选择，并在仿真实验中验证了其有效性，今后将在实际的视觉系统中进行构建。

The purpose of this study is to construct the visual model for efficient visual information processing, which has "glance view" and "searching view" which are human's visual sense process stages. Then those process stages are chosen properly as an index which used the effective visual field that is a human's psychological concept.Moreover, the psychological experiment was carried out about the effective visual field in human visual function, and its characteristics were analyzed. The new system is constructed by applying the model with effective visual field to the conventional vision system. The simulation experiment was carried out, and the effectiveness of this model was shown.Then, the psychological experiment about the effective visual field was carried out for various patterns. It was confirmed that there was the relation of the trade-off between the complexity of the visual processing problem and effective visual field. And it was shown that the human is quantifying information from the externality by some methods in a moment. Then, we tried the construction of the high-level vision system by including such features in the system, can choose the visual direction which facilitates the 3-D object recognition.It is difficult to measure the effective visual field as an index. Therefore, the object image entropy, which is an index that shows the information amount of the object image, was used as the index instead of it. The object image entropy is one-dimensional scale simply calculated from the object image. In the psychological experiment, it was confirmed that it was easy to be recognized, if the entropy is high. In addition, the effectiveness was confirmed in the simulation experiment that automatically chose visual direction by modeling the 3-D object in the computer.In the future, it is constructed in actual vision system.