Quality estimation of product group from nonuniform texture images by selective self-inhibited learning
✍ Scribed by Kyouji Tanaka; Yoshiaki Shirai
- Book ID
- 102661573
- Publisher
- John Wiley and Sons
- Year
- 2000
- Tongue
- English
- Weight
- 659 KB
- Volume
- 31
- Category
- Article
- ISSN
- 0882-1666
No coin nor oath required. For personal study only.
✦ Synopsis
In most of the classification methods for images, only the effective information is utilized, which is extracted from the original image using a priori knowledge. In contrast, this study intends to estimate the quality of the product group from the nonuniform texture image, without using a priori knowledge and using only the result of decision by the expert as supervisor data. A neural network (NN) is used for classification. It is shown first that the classification based on the principal components, which is a linear classification procedure, is difficult. Then three learning methods are considered. In the first method, the small regions of the image are input and the frequency filter is formed by learning in the NN. In the second method, the wavelet component is used as the input. In the third method, the wavelet component is used as the input, and the learning is selectively self-inhibited according to the intermediate response to the small region. In an experiment, it is shown that the first two methods require a long time for learning, although the classes can be formed, and the third method is effective.