Dynamic Processing of Visual Patterns

为了开发下一代视觉信息处理系统的新设计原则，我们集中研究了生物大脑视觉系统中的主动和动态过程。我们使用建模的方法来揭示大脑的机制，并设计出先进的视觉模式识别系统。我们并行开展了多项研究，取得了以下成果。能够从复杂背景中识别人脸的神经网络模型。它可以集中注意力并将面部成分(眼睛和嘴巴)从识别的人脸中分离出来。能够正确识别部分遮挡模式的神经网络模型。用于在现实世界中识别手写数字的新版本的新认知电子设备。我们之前开发的新认知电子管是一种模式识别系统，其体系结构是从哺乳动物的视觉系统中提出的。当我们使用3000个字符进行训练时，识别率随着训练集大小的不同而不同，达到了98%以上。一种新的学习规则，具有移位不变感受野的细胞根据该规则自组织。根据这一学习规则，在网络中产生与初级视皮层中的简单细胞和复杂细胞类似的细胞。可以记忆和回忆空间地图的神经网络模型。该模型模拟了一个人在二维空间中走动时记忆和回忆空间地图的情况。该模型记忆的是零散的地图，但通过一个连续的链条回忆过程，可以无缝地检索到覆盖大范围的图像。立体算法，从眼睛间不成对的点提取深度线索。

Aiming to develop new design principles for visual information processing systems of the next generation, we have concentrated our research on the active and dynamic processes in the visual system of the biological brain. We used modeling approach to uncover the mechanism of the brain and to design advanced systems for visual pattern recognition. We have performed various researches in parallel and have obtained the following results.1. Neural network model that can recognize faces from complex backgraound. It can focus attention to and segment facial components (eyes and mouth) from the recognized face.2. Neural network model that can recognize partly occluded patterns correctly.3. Neocognitron of a new version for recognizing handwritten digits in the real world. The neocognitron, which we have developed previously, is a pattern recognition system whose architecture has been suggested from the mammalian visual system. The recognition rate, which varies depending on the size of training set, was over 98% when we used 3000 characters for the training.4. New learning rule by which cells with shift-invariant receptive fields are self-organized. With this learning rule, cells similar to simple and complex cells in the primary visual cortex are generated in a network.5. Neural network model that can memorize and recall spatial maps. The model emulates a situation where a person memorizes and recalls spatial maps when he moves around in a two-dimensional space. The model memorizes fragmentary maps, but can retrieve an image covering a wide area seamlessly by a continuous chain process of recalling.6. Stereo algorithm that extracts a depth cue from interocularly unpaired points.