Welding Defect Detection and Classification Using Geometric Features

In this paper we present a welding defect detection system using radiographic images. Main goal is to craft a dependable system because a human evaluator is not a stable evaluator besides other humanoid constraints. We present a novel technique for the detection and classification of weld defects by means of geometric features. Firstly noise reduction is done as radiographic images contain noise due to several effects. After this we tend to localize defects with maximum interclass variance and minimum intra class variance. Further we move towards extracting features describing the shape of localized objects in segmented images. Using these shape descriptors (geometric features) we classify the defects by Artificial Neural Network.