A quantitative model of width measurement for irregular crack is established. A novel method on distinction between true and pseudo cracks is presented, which uses the theory of edge detection and morphological dilation operation. By selecting two approximate equivalent sets of target pixels and background pixels at the marginal areas of cracks′ both sides, the best threshold is calculated adaptively. Then the edge′s pixels and the movement of crack through the binarization image are extracted. Using the quantitative model, the width of crack can be measured. The experimental results show that the proposed method can effectively overcome the image noise, manipulate any irregular crack, and also remove the pseudo crack. The repeatability precision of the method is superior to 0.1 mm.