The Utilization of Image Subtraction and Wavelet Decomposition-Reconstruction for Improving FCM Based Segmentation of Radiographic Weld Defect

Automatic weld defect segmentation is an important step in radiographic weld defect identification system. Fuzzy C Means (FCM) clustering, which had been proven capable and applied for many images segmentation applications, was utilized as the segmentation basis of weld defect object. In this study, a method using Laplacian sharpening of image subtraction and 2-D wavelet decomposition-reconstruction is utilized for improving the result of FCM-based weld defect segmentation. Image subtraction was applied by subtracting the background image, which was estimated by 30×60 average filtering, with the origin image and followed by image sharpening operation using Laplacian filtering to reduce the background intensity and to enhance the foreground area. Two level decomposition of 2D db4 wavelet was employed to produce wavelet coefficients, then 2 levels wavelet reconstruction was performed only using approximation coefficient to synthesize a new image. In the last step, reconstructed image was segmented by FCM clustering to obtain a weld defect image. The proposed method was tested on five different weld defect types sample and it was evaluated using mutual overlap approach. The evaluation result showed that this method increase the mutual overlap metric which obtained 57.46%.

Keywords: Radiographic weld defect segmentation, fuzzy c means, image subtraction, wavelet decomposition-reconstruction