Crack detection is crucial for assessing structural safety. Traditional image processing methods for crack detection in tunnels often suffer from high noise levels and low accuracy due to uneven lighting and severe noise pollution. To address these challenges, this study proposes a tunnel crack detection algorithm based on machine vision. First, tunnel images are filtered in the frequency domain and differentiated in the spatial domain to enhance texture features. Then, image segmentation is performed with area parameter T_v, saturation parameter T_s and special parameters Tv' and Ts' to remove background noise and reduce misdetections, facilitating complete crack detection. Finally, a lightweight crack-connection algorithm is designed to bridge discontinuities in crack images, based on the stability and development pattern of cracks in this application scenario. Experimental results show that the proposed method effectively extracts complete cracks, achieving an image recognition accuracy of 94%, and a recall rate of 98%, meeting the requirements of practical engineering applications.