MDS-SIFT: An improved SIFT matching algorithm based on MDS dimensionality reduction
Zhiping Zhou, Cheng Shimeng, Zhongmin Li · 2016
As a result of high dimensional feature descriptor, the image matching algorithm based on SIFT feature requires expensive computation. Since only the local gradient information is considered when computing the feature descriptor, it produces considerable mismatching points. In order to overcome above deficiencies, an improved matching algorithm is presented in this paper. At first, the dimensions of feature descriptor are reduced by MDS (multiple dimensional scaling) algorithm based on variance explanation ability as like principle component analysis. Then, an improved bidirectional matching strategy is proposed based on the consistency of characteristics and distance ratio. Finally, the local texture character of each matching pair is analyzed to reduce some mismatching points. The simulation results show that the newly presented algorithm not only has better matching rate than other similar algorithms but also decreases the executing time.