Abstract: In order to solve the problems of poor anti-interference ability and low recognition accuracy of traditional crack identification algorithms，an image identification system for concrete surface existing cracks was established by improving the existing seed filling algorithm to achieve accurate extraction of crack information.In the pre-processing，the illumination uneven coefficient screening algorithm was proposed to quickly screen the image，which improves the efficiency of uniform light processing.In the crack identification，the existing seed filling algorithm was improved，so that it can automatically determine the growth point of the crack.And the image segmentation of the crack was realized by combining the eight-direction search method with the boundary judgment of the relative threshold method.Then a series of complex background interference were eliminated with the connected domain filter.In the feature extraction stage，by introducing morphological processing，burr removal and node Euclidean distance，the quantitative information of the number of cracks，length and width was accurately obtained.Compared with traditional concrete crack image recognition algorithms，this algorithm achieves a greater degree of unification of efficiency and accuracy，and the extraction error of the crack length and width values is controlled within 10%.

Key words: improved seed filling algorithm, concrete crack, crack identification, image segmentation, feature extraction, identification accuracy