The contrast level of the radiographic image sequence risks to be inconstant in the detection procedure due to the variation of the detection conditions. The adaptability of the detection method is required. A varying contrast level brings the problem to the segmentation of the potential defect region. To study how the proposed method can handle with the unstable performance of the segmentation, we consider the case equivalent to applying varying threshold to the same image sequence.
The threshold varies from 90 to 130 to test how the proposed Kalman filtering detects true defects that appear in the image sequence. The low quality of the images brings severe problem of detection. Fig. 4 shows the miss detection rate and the false alarm rate of the potential defect regions, before applying the Kalman filtering method. The miss detection rate and the false alarm rate are both sensitive to the change of the threshold. The threshold window [116,124] allows the two rates under 50%. The threshold window [117,122] allows the two rates both under 45%. The optimal case appears when the threshold equals 119: the miss detection rate is 38.5% and the false alarm rate is 36.4%. Obviously the detection result is not satisfactory.
With the application of the Kalman filtering method, the true defects can be successfully distinguished from the false alarms. The Kalman filtering method tracks the continuous motion of the true defects, and thus the false alarms are avoided. For the detection of defect, we can group a number of successive appearances that belong to one true defect. The detection of the true defects in the whole image sequence is verified under different threshold of potential region segmentation. Fig. 9 shows the region of threshold in which each weld defect can be detected. In the threshold range of [101,119], 100% detection rate and 0% false alarm rate can be achieved.