Grouping Beamlet Transform for Surface Crack Detection

Abstract

Beamlet transform is an excellent multiscale geometric analysis method. It has a great capacity of extracting line features from images under noise. However, it is too slow since a mass of redundant beamlets waste much time. In fact, only several of them are helpful in many applications. In this paper, grouping beamlet transform is presented to fasten line feature detection for surface crack detection. Geometric flows used in grouplet transform are introduced to determine geometric structures of an image. In each recursively partitioned box, only the beamlets along with major direction are generated. So that lots of useless beamlets for the image can be excluded from the following integral computation. Experiments on various optical images show that grouping beamlet transform is able to detect line features in an image the same as classic beamlet transform, however, the former runs much faster than the latter on any of the tested images. In some cases, only about 60% of consuming time in classic beamlet transform may be needed in grouping beamlet transform. As an improvement of beamlet transform, grouping beamlet transform will be more applicable in the field of surface crack detection.