We collected real crack image data to test and analyze the image-based crack detection results. We used 18 images with 1920×1080 resolution sequentially numbered from 1 to 18. And we performed labeling operation to classify the crack image into crack and background regions. online for free use. For the selection of a crack region, we applied a simple rule of extracting a crack edge and selecting the segment matching the true value. The inner region of the selected crack edge (i.e.