As a result of the influence of surface characteristics, it is difficult to identify cracks on complex surfaces using the conventional processing of a single-frame image that misses important infrared thermal image sequences. In this paper, the red-green-blue normalisation method and the interframe difference method are used to make full use of the sequences and to characterise infrared thermal images of cracks on complex surfaces. An experiment to prove the effectiveness of the detection method was carried out on weld steel and carbon fibre-reinforced polymer (CFRP) steel plate. The selected infrared thermal image was divided into an objective area and a background area according to the relationship between three parameters for red-green-blue and objective temperature. The targeted and interfering information of the objective area were separated and processed based on the interframe difference method. The index of crack detection was extracted from the interframe difference matrix, which was constructed from two selected infrared thermal images. This method takes into account the local and global spatial information of the image and the maximum crack length measurement error was approximately 6%.@en