Research on Fabric Image Mosaics Based on SURF

The SURF algorithm is widely used because of its high registration accuracy. However, because the algorithm detects the feature points of the entire image, it cannot meet the characteristics of single texture, similarity and density of fabric images. At the same time, because of the long search time and low computational efficiency of image registration using KNN method for feature detection and matching, this paper studies the image registration algorithm based on KD tree search and SURF features. The input of the algorithm is to register the set of feature points in the image, and generate a feature description vector, and then establish the KD tree index of the feature description vector. When performing image registration, the ratio method is used as the criterion for matching feature points. Each feature point in the registered image is regarded as a query point, and the first two feature points closest to the query point are found in the KD tree. Finally, the MASC algorithm combining geometric motion parameter estimation and interpolation is used to purify the matching points. The experimental results show that the KD tree-based SURF algorithm, which combines the MASC algorithm to eliminate mismatched point pairs, is better than other algorithms in terms of speed and accuracy in feature detection.