Systematic design of object shape matched wavelet filter banks for defect detection

In our previous works, we have presented methods for optimizing wavelet filter banks, which can be used for classification of image objects. The wavelet filter banks were designed to be biorthogonal, which enables a multiscale analysis on given image data. Moreover, the filters were optimized with respect to the shape, which helps the filter banks to inherit the property of the objects. This optimization is only possible with the help of so called object filters designed to have the curve of typical objects of each class. In contrast to previous works where object filters were designed manually, a systematic and automatic design method for object filters is introduced in this paper. The new designed filters were used to optimize wavelet filter banks for classification problems. The evaluation of this method was done by comparing the results with the ones of wavelet filter banks based on the previously used object filters.