The current graphic retrieval algorithm based on CSS (Curvature-Scale-Space) only used the peak value of curvature space image which varied according to the image shape and neglected the important features of graph, which caused lower retrieval accuracy and efficiency. In order to solve the problems, the image retrieval algorithm was put forward based on 2D Fourier transform coupled with improved curvature scale space (CSS). Firstly, it considered the dynamic change in curvature zero crossing point’s process, defined the peak threshold control criterion, and improved the CSS mechanism by combining parabola fitting technology to remove the pseudo-peak point, and take into account the non peak point information of image shape and get CSS parabola fitting image. Introducing the 2D Fourier transform, it substituted image shape with CSS parabola fitting image and achieved 2D Fourier transform of curvature scale image. Finally, it conducted normalization, built the curvature-Fourier descriptor, and constructed the Euclidean distance between the query image and the image database, and finished graphic retrieval. Database test results showed: compared with the current technology based on curvature scale space and 1D Fourier descriptors, the new algorithm had higher retrieval precision and efficiency as well as better PR curve. It can further improve the precision and efficiency of graphic retrieval and have good application value in trademark retrieval, etc.