Noise reducing and image processing for debonds quantitative identification by infrared thermography

An adhesive structure specimen, fabricated with several back-drilled flat holes of different size to simulate debond defects, was inspected by infrared thermography. Three larger defects could be seen directly by the raw images. However, two smaller defects could not be detected by the raw images because of high noise. The noise contributed by the variation of surface emissivity, was eliminated to some extent by coating the testing surface with uniform emissivity black paint in the experiment. It is found that the surface temperature changes with time satisfying exponential form. Therefore, a robust reducing noise method - exponential fitting of every pixel as a function of time, was presented to eliminate unwanted temporal noise in the raw images. Spatial noise was reduced by the combination some standard filtering algorithms e.g. average filtering with high-pass filtering, which made the invisible defect be detected. Finally, image segmentation algorithms based on watershed and 3D tomogram were used to visualize the defects, and defect´s parameters were also estimated by the image after segmentation. The results show that the methods used in this work could improve defect detectability and achieve defect quantitative identification.