Objective In response to the sensitivity of the scale invariant feature transform SIFT algorithm to noise during the three-dimensional reconstruction process leading to errors in feature point extraction and matching as well as long runtime an improved SIFT algorithm was proposed to enhance the accuracy of feature point extraction and reduce runtime. Methods The improved SIFT algorithm first traversed the pixels of the image. For each target pixel it compared the grayscale values with those of its eight neighboring pixels. If the difference in grayscale values between adjacent pixels and the target pixel was less than a specified threshold the adjacent pixel was marked as a similar point. Based on the number of similar points whether the target pixel was an interest point was determined. If the number of similar points met specific conditions the target pixel was determined as an interest point and then the SIFT algorithm was used to extract feature points within the region centered on the interest point. Results In experiments comparing different threshold settings and images of different sizes the results indicated that the improved SIFT algorithm achieved an approximate 10% increase in feature point extraction accuracy and saved around 25% of runtime compared with the traditional SIFT algorithm. Conclusion The experimental results demonstrate that the proposed improved SIFT algorithm effectively enhances the quality of feature point extraction by introducing noise suppression and interest point filtering reducing the error rate in feature point extraction and matching and significantly reducing runtime.