Automatic Road Pavement Assessment with Image Processing: Review and Comparison

Figure 9

The evaluation criteria. In this figure, two segmentations of the same crack are represented the black line corresponds to a manual (or reference) segmentation, and the red line is an estimated segmentation. The goal is to evaluate the quality of the estimated segmentation that corresponds to the positives (P). All the nonselected pixels that do not correspond to the crack are called the true negatives (TN). The piece of line with the two colors (red and black) are the correct detections or true positives (TP). In this table, different criteria are introduced, but, in this work, we have used the DICE because this coefficient well represents what we want to measure: the quality of the detection against the percentage of the crack that is detected, in order to determine how to reduce false detections whilst increasing the density.