A image registration method based on SURF and BRIEF

Considering SURF feature description having great computational complexity and feature matching not having high accuracy, an advanced image registration method is proposed. At first it uses SURF to extract the feature points from two images, then calculates Harris response values of every feature points. The feature points of low quality are canceled. Then BRIEF descriptor is utilized to characterize those feature points. Lastly the improved K-means algorithm is used to classify the data, and the classes with more matching points are reserved. The advanced RANSAC algorithm is used to compute the transform matrix between the two images. The experiments verify the algorithm's efficiency.