Detecting linear features in weld seam images based on beamlet transform

Weld seam images are replete with noise and ordinary line detection algorithms such as Hough transform is not an ideal approach to extract significant linear features from it. We present a novel line detection algorithm based on beamlet analysis. The algorithm is described in detail after introducing the beamlet dictionary and beamlet transform. Taking into account of some special characteristics of welding image processing, we add an orientation-thresholding step to the standard beamlet-based line detection algorithm. Experiments show that our line detection algorithm is powerful in anti-noising and is especially suitable to detect linear features in heavily noisy weld seam images. We directly extracted one significant linear feature of the weld seam from both the original experimental image and the extra-noised images (SNR ges6 db) exactly and efficiently at scale 0 with the help of orientation threshold. Neither pre-processing nor post-processing of the image was performed. This implies that our algorithm can dramatically improve the efficiency of weld seam image processing.