As an essential part of the power transmission, the quality of the cable core is important for power safety and transmission efficiency. However, the knife mark texture in the cross-sectional image of the cable core affects the image segmentation effects seriously, which could reduce the measurement accuracy while using visual measurement for cable structural parameter detection. To solve this problem, this article proposes an improved frequency domain filtering to eliminate the knife mark texture of cable core crosssection image. Firstly, the cable core cross-section images are decomposed by the relative total variation (RTV) model to obtain the texture image. Then, the texture image is processed by fast Fourier transform to achieve the main energy direction of the texture. Finally, the Butterworth notch filter with oval-shaped neighborhood narrow stopband is designed to reduce the energy of the corresponding band-shaped area. Experimental results show that the elimination rate of knife mark texture is 92% , which proves the effectiveness of the proposed method.