2-dimensional color sensor with combined neural network
Mutsunori Takahashi, H. Hashimukai, Hideyuki Ando · 1991
Summary form only given. A 2D color sensor has been developed to identify objects using color image data. When using this sensor the user must set several parameters for each object and this setting procedure is difficult using conventional methods. An approach to automatic parameter setting for identification by using a combined neural network has been developed. This combined neural network is composed of two network models. The method extracts the object's feature distribution in red-green space using the Kohonen feature map. Feature vectors are made from these distributions. The feature vectors are used to train a multilayer network by the back-propagation method for identification.>