Wood surface defects can seriously affect the quality, performance and use value of wood; therefore, the detection of wood surface defects is beneficial to improving the utilization of wood, saving the existing wood resources and easing the shortage of forest resources. So it is important to study the method of image segmentation of wood surface defects. Aiming at the shortcomings of the traditional C-V (Chan-Vese) model which cannot segment gray images, we used a combination of C-V model and the morphological method for a comparison with sole C-V model algorithm. Given the weakness of the traditional C-V model or combined with the morphological method, the local fitting function and the Gauss kernel function which are based on the C-V model are introduced. Thus, an improved algorithm based on C-V model is proposed which overcomes the shortcomings of the C-V model. Targeting a single wood surface defect, the segmentation of images has been performed by three algorithms, i.e., the C-V model algorithm, combination of C-V model and morphological method, and our improved algorithm, for a contrast test. The test shows that the C-V model is capable of segmenting the images of wormholes and slipknots, but it is difficult to segment the images of encased knots. With the morphological method, the small holes and noise after segmentation can be effectively eliminated, but it is difficult to split out the defect image segmentation of encased knots completely, and still cannot resist interference of dead knot images of wood texture itself. The segmentation of defect image based on our improved C-V model algorithm is quicker and more accurate, and can reduce the number of iterations, shorten segmentation time and make the segmentation contour more smooth and complete. Our study shows that the improved algorithm is able to finish the multi target experiment, and therefore it is feasible, superior and practicable.