A new improved level set algorithm is proposed to solve the problem of initial contour sensitivity in cervical image segmentation and the problem of unclear image gray level. Firstly, the image is denoised by anisotropic filtering algorithm. Then the region growth algorithm was used on the binary image to extract the rough cervical lesion area. Finally, a level set model based on the new symbolic pressure function is established to refine the initial segmentation. The algorithm can combine local information with global information and automatically allocate the proportion between them. The accuracy, sensitivity and specificity of this method can reach 81. 11%, 63. 97% and 78. 64% respectively, which are 30. 69%, 15. 15% and 4. 37% higher than the traditional level set algorithm. Therefore, this improved level set algorithm has some value and significance in practical application.