직/편성물 염색

Color Recipe Prediction by Artifical Neural Networks

  • 출판일1997.09
  • 저자
  • 서지사항
  • 등록일 2016.11.02
  • 조회수 294
Researchers investigated a neural network alternative to the conventional two-flux theory of Kubelka and Munk for computer colorant formulation. An experiment used a multilayer feed forward network model to train several nets of different topologies via the Kubelka-Munk equation for reflection of opaque diffusers, then determined optical material parameters of a vendible automated repairing system. Researchers extended the number of colorants and performed calculations on reflection rather than color space. Results using different strategies for pattern generation demonstrated that feed forward networks are trainable to small colorant assortments. Training error deteriroated significantly with increasing numbers of colorant components. Appropriate filter mechanisms redeuced the number of master training and test patterns while improving network performance. 29 refs.
이전글 다음글 목록