Abstract: By employing pre-processing of morphologic enhancement and area-reconstruction,a novel method for image segmentation was presented based on watershed transformation.And using the method,ultrasonic test image for a joint interface between dissimilar materials was segmented.To validate the reliability of this method,the sample was destructed and tested.The results show that over-segmentation caused by traditional watershed transformation can be avoided by pre-procession of morphologic enhancement and area-reconstruction.Meanwhile,the test image can be blocked adaptively according to the distribution feature of defect and then the defect can be extracted and quantified by threshold segmentation.Destructive experiment shows the presented method is very efficient with high reliability.