抄録
Gray scale image analysis was applied to evaluate visual textures of 29 kinds of lace patterns. The mean gray level value was obtained from the gray level image. The angular second moment, contrast, correlation and entropy extracted from the gray level co-occurrence matrix were measured as textural feature parameters. The fractal dimension was determined from the fractal analysis of the relief of the curved surface of the gray level image. These image information parameters are useful for the visual evaluation of lace patterns. In this study, the identification of a visual evaluation system using neural networks was discussed. A high performance neuron training algorithm using a Kalman filter was introduced to tune the network in order to maximize the accuracy of the description of the visual evaluation system. The Kalman filter neuron training algorithm gives much better results than other neuron training algorithms. The trained neural network model was successfully implemented to show the feasibility of neural network applications for visual evaluation.
