When defect detection is carried out on the surface of liquid crystal screen, the horizontal and horizontal texture of the surface will interfere with the detection accuracy. In this paper, a defect detection method for liquid crystal screen is proposed. Firstly, the texture of liquid crystal screen is suppressed by the image structure extraction method based on the relative total variation model. This model is applicable to both non-uniform and anisotropic textures, and can effectively decompose the structural information and texture in liquid crystal image. The extracted structural information is enhanced by high-boost filter while keeping the characteristics of low-frequency component, and the saliency detection of defects in the image is carried out. The experimental results show that this method can effectively suppress the background texture information and retain the defect information completely, which can realize the saliency detection of LCD screen defects and improve the accuracy of LCD image defect detection.