The work aims to improve the matching speed and accuracy of product packaging pattern images and obtain the key factors that affect these two aspects in this algorithm. By using Gauss Pyramid to sample down the original images, the images were converted to a scale space representation. Then speed up robust features (SURF) method was adopted to extract feature points, and Brute-Force algorithm was used to match these points. Finally, after sorting the Euclidean distance of the matching points in each group, the mismatching points were eliminated and a certain quantity of correct matching points remained. The influence of Gauss Pyramid, Hessian value and color space on the calculation speed and the number of the extracted feature points were obtained. The change of color space did not affect the number of feature points, and the difference of the computation time could be ignored. When the scale space was reduced by 1/2, the computation speed could be improved by 75%, and 2/3 of the redundant feature points were eliminated at the same time. The method in the paper can effectively improve the speed of feature extraction, and is characterized by high matching precision and strong robustness. Meanwhile, it has strong adaptability to the change of rotation angles.