Feature recognition for underwater weld images

Real-time sensing and detecting of underwater weld position is a key technique. Laser vision sensing is a good-prospect detecting method. Therein welding image processing and feature recognition are important parts. Noise features of underwater weld image in different water conditions are described. Underwater V-groove weld image pre-processing is discussed. Mean Shift algorithm application to underwater weld image segmentation is studied, and Hough transform to recognize image features of underwater weld is explored. Experiment results show, after such a series of operation as power transformation, limited contrast histogram equalization, top-hat operation, omnidirectional structuring element cascade filtering, underwater weld image is well pre-processed; weld feature image is more effectively segmented by Mean Shift algorithm than by C-means clustering; Hough transform is applicable to precisely recognizing V-groove weld feature points.