Enhancement Of X-Ray Images For Industrial Applications

Abstract

In metallic structures, defects are usually manifested as cracks, surface wear, poor weld spots and slag inclusions. The timely detection of such faults is useful in preventive maintenance routines. Various researchers have shown that manual inspection on structures is unreliable, because the human visualization process is susceptible to inaccuracies and can be biased. Hence, manual results lack reliability and consistency. This hampers with the entire process of preventive maintenance. Similarly, manual inspection is costly and consumes a lot of time. This has given way to the development of numerous automated techniques for defect detection. This thesis is developing a simple and efficient technique for automatic defect detection on digital radiographic images for industrial application. Otsu’s and percolation thresholding and segmentation algorithms have been used in developing the proposed algorithm. In addition, it is desirable that images should be in a form in which information can be extracted easily, interpreted and where necessary permit for further image processing. In this thesis various image enhancement techniques have been reviewed so as to improve on the contrast of digital radiographic images obtained from the industrial sector. These techniques involve diverse procedures such as histogram equalization, contrast stretching, edge sharpening and noise reduction. These processes are achieved by using both the frequency and spatial domains. This investigation involves the use of both techniques and aim at improving existing algorithms. One of the envisaged approaches is the use of histogram based contrast enhancement supplemented by some wavelet based noise suppression technique. This eliminates low contrast and improves the degraded features. MATLAB based computer simulations have been applied in determining the efficacy of the proposed algorithms. The radiographic images used were obtained from Quality Inspectors Limited, Kenya Bureau of Standards and from various internet sites which contain images used by other researchers. A total of sixty two images and sixteen templates were used. Both subjective and objective techniques have been applied to determine the usefulness of the suggested techniques. The performance of these algorithm, which involved image denoising using wavelets, image enhancement to improve on the image contrast and automatic defect detection, are better in comparison to existing defect detection techniques and the outcomes achieved were remarkable in regard to the rate of defect detection and the defects identified such as cracks, weld pores and slag inclusions.