The goal of image fusion is to integrate data from different images of the same scene. The outcome of image fusion is a new image that is more suitable for both human and machine observation, as well as for further image processing tasks. Thermal images often suffer from noise issues such as non-uniform radiation and non-uniform emissivity, which can overshadow the subtle thermal contrast of deeper defects. Consequently, various signal processing methodologies have been employed to minimize noise and delve into deeper depth details. However, no single method has proven superior in providing comprehensive details, necessitating the merging of information obtained from different processed images using various fusion algorithms. To achieve more precise subsurface details, we are interested in applying suitable fusion techniques to processed thermal images and then combining them into a new single image. This paper proposes a non-subsampled Contourlet transform-based data fusion approach to incorporate all subsurface details into a single image, enabling precise subsurface analysis. The effectiveness of this approach has been verified through experiments conducted on a carbon fiber-reinforced plastic specimen containing embedded flat-bottom holes, employing various processing methods.