Quality monitoring of steel surface using wavelet packet transfrom

Product quality is often reflected by visual appearance of a product quality. The visual appearance of a product surface was measured by lab technician. However, a critical problem for the measurement of quality is that such measurements are time-consuming and costly. Nowadays, surface quality can be monitored by digital image sensors and automated image grading system. For image grading system, feature extraction from images is an essential step in characterizing quality of a product surface. In previous works, wavelet texture analysis (WTA) based on discrete wavelet transform (DWT) has been recognized as one of the most successful feature extraction methods for classifying steel quality. In this paper, different types of wavelet transforms (wavelet packet transform and the discrete wavelet transform) are compared in classification of surface quality of rolled steel sheets. It is shown that the performance of the wavelet packet transform is superior to that of the discrete wavelet transform in terms of classification performance as well as Fisher´s criterion. The characteristics of the image data reveals that due to its equal frequency bandwidth, WPT is more suitable in extracting textural features in cases where textural information of different classes of images is not confined within a certain (spatial) frequency regions.