Abstract: Defect size measurement has always been a popular topic in infrared thermography. Typically, images containing all defects are used for defect size measurements, which leads to significant measurement errors. We propose using clear images corresponding to the defects for measurement. First, the correlation function method was employed to filter out clear images corresponding to defects at different depths. Then, the half-width measurement algorithm was applied to achieve automatic defect size measurement. By measuring plastic specimens with defects of varying depths, it was shown that for a 20 mm defect, using images containing all defects resulted in a maximum measurement error of 12% and an average absolute error of 6.1%. Using clear images selected by correlation functions for defect size measurement reduced the maximum measurement error to 6% and the average absolute error to 2.6%. This method can effectively improve the accuracy of defect size measurement.