公开(公告)号：JP2008102589A

公开(公告)日：2008-05-01

申请号：JP2006282496

申请日：2006-10-17

Applicant: Toa Corp ,  ティーオーエー株式会社

Inventor： MATSUI NOBUYUKI ,  KAMIURA NAOTAKE ,  ISOGAWA TEIJIRO ,  OGAWA YUZO ,  OTSUKA AKITSUGU ,  IWATANI KENJI

IPC: G06T7/00 ,  G06N3/00 ,  H04N5/232 ,  H04N7/18

CPC classification number: H04N5/262 ,  G06T7/277 ,  H04N19/537

Abstract: PROBLEM TO BE SOLVED: To provide a novel moving image processor detecting a moving object in an moving image using a self-organizing map. SOLUTION: A composite video signal output from a camera 20 is converted into color image data by an input conversion part 50. The color image data is input to a feature extraction part 58 through an image dividing part 52 and a frame setting part 56. The feature extraction part 58 extracts n-dimensional features of the input color image data, and the extracted feature data is input to a control part 60. The control part 60 constitutes a block unit learning type self-organizing map together with a map 62, and identifies whether each pixel forms a moving object region or a background region by applying the feature data to the map 62. Based on the identified result, the control part 60 controls an output conversion part 70 to display only the moving object region on a monitor 40. COPYRIGHT: (C)2008,JPO&INPIT

Abstract translation: 要解决的问题：提供使用自组织图来检测运动图像中的移动物体的新颖的运动图像处理器。 解决方案：从相机20输出的复合视频信号通过输入转换部分50转换为彩色图像数据。彩色图像数据通过图像分割部分52和帧设置部分被输入到特征提取部分58 特征提取部分58提取输入彩色图像数据的n维特征，并将所提取的特征数据输入到控制部分60.控制部分60构成块单元学习型自组织图与地图 并且通过将特征数据应用于地图62来识别每个像素是否形成移动物体区域或背景区域。基于所识别的结果，控制部分60控制输出转换部分70以仅显示移动物体区域 监视器40.版权所有（C）2008年，JPO＆INPIT

公开(公告)号：JP2006140952A

公开(公告)日：2006-06-01

申请号：JP2004330983

申请日：2004-11-15

Applicant: Nobuyuki Matsui ,  Sutetsupu One:Kk ,  Toa Corp ,  ティーオーエー株式会社 ,  伸之 松井 ,  株式会社ステップワン

Inventor： MATSUI NOBUYUKI ,  ISOGAWA TEIJIRO ,  SODO HIROMI ,  MAEDA KAZUAKI ,  OGAWA YUZO

IPC: H04N1/46 ,  G06N3/00 ,  G06T1/00 ,  H04N1/60

Abstract: PROBLEM TO BE SOLVED: To obtain a processing result as expected corresponding to the luminance of color image data, in an image processor which processes the color image data by a neutral network.
SOLUTION: The color image data Do are inputted to the network part 16 of the neutral network 14 of a BP-method learning type after normalized to values of 0 to 1 by a normalization part 12. The network part 16 applies luminance conversion processing to the color image data Do on the basis of parameters Wnm, Wpn, θn and θp set from a parameter setting part 18. At this time, the color image data Do are processed as a quaternion. Namely, three components, an R component, a G component and a B component constituting the color image data Do are collectively processed as three-dimensional vector data. Furthermore, the parameters Wnm, Wpn, θn and θp set in the network part 16 are selected according to the luminance of the color image data Do.
COPYRIGHT: (C)2006,JPO&NCIPI

Abstract translation: 要解决的问题：在通过中性网络处理彩色图像数据的图像处理器中获得与彩色图像数据的亮度相对应的预期的处理结果。 解决方案：通过归一化部分12将彩色图像数据Do输入到BP方法学习型中性网络14的网络部分16，其归一化为0至1的值。网络部分16应用亮度转换 基于从参数设定部18设定的参数Wnm，Wpn，θn，θp对彩色图像数据Do进行处理。此时，彩色图像数据Do被作为四元数进行处理。 即，将构成彩色图像数据Do的R成分，G成分和B成分的三个成分统称为三维矢量数据。 此外，根据彩色图像数据Do的亮度来选择网络部分16中设置的参数Wnm，Wpn，θn和θp。 版权所有（C）2006，JPO＆NCIPI

公开(公告)号：JP2005078175A

公开(公告)日：2005-03-24

申请号：JP2003304745

申请日：2003-08-28

Applicant: Nobuyuki Matsui ,  Sutetsupu One:Kk ,  Toa Corp ,  ティーオーエー株式会社 ,  伸之 松井 ,  株式会社ステップワン

Inventor： MATSUI NOBUYUKI ,  ISOGAWA TEIJIRO

IPC: G06N3/04

Abstract: PROBLEM TO BE SOLVED: To provide a data processor for processing three-dimensional data by a neural network to acquire expected processing results.
SOLUTION: Three-dimensional data x
m are normalized into the values of 0 to 1 by a normalizing part 12, and inputted to a network part 16 of a BP learning type neural network 14. The network part 16 operates predetermined processing to the three-dimensional data x
m based on parameters w
nm , w
pn , θ
n and θ
p set by a parameter setting part 18. At that time, the three-dimensional data x
m are processed as quaternion. That is, three components configuring the three-dimensional data x
m are processed in a batch as three-dimensional vector data. Thus, it is possible to maintain the relevancy of those three components. Then, the processed data y
p by the network part 16 are outputted to the outside as the output data of a data processor 10.
COPYRIGHT: (C)2005,JPO&NCIPI

Abstract translation: 要解决的问题：提供一种用于通过神经网络处理三维数据以获取预期处理结果的数据处理器。 解决方案：归一化部分12将三维数据x m 归一化为0至1的值，并输入到BP学习型神经网络14的网络部分16。 网络部分16基于参数w nm ，w pn ，θ n 对三维数据x m / SB>和由参数设置部分18设置的θ p 。此时，三维数据x m 被处理为四元数。 也就是说，构成三维数据x m 的三个组件以批量处理为三维矢量数据。 因此，可以保持这三个部件的相关性。 然后，网络部分16的处理数据y p 作为数据处理器10的输出数据输出到外部。（C）2005，JPO＆NCIPI