Specific features for the analysis of fringe images

In optical nondestructive testing, a novel solution is presented for fault detection based on the interpretation of fringe images. These images can be acquired using different optical methods, such as structured lighting or interferometry. We propose a set of eight special features adapted to the problem of surface inspection using structured illumination. These characteristics are combined with six further features specially developed for the classification of faults using interferometric images. We apply two kinds of decision rules: the Bayesian and the nearest neighbor classifiers. The proposed features are evaluated using a noisy and a noise-free image data set. All patterns were obtained by means of structured lighting. Concerning the noisy data set, we obtain better classification rates when all the 14 features are used in combination with a one-nearest-neighbor classifier. In case of a noise-free data set, we show that similar classification rates are obtained when the 14 features or only the 8 specific features are involved. The methods described are designed to address a broad range of optical nondestructive applications involving the interpretation and classification of fringe patterns.