2016 IEEE Advanced Information Management,Communicates,Electronic and Automation Control Conference (IMCEC)

119 / 2016-06-18 18:44:30

An Unsupervised Feature Learning Approach to Feature Description

Image Registration,feature Description,Machine Learning,Unsupervised Feature Learning

In this paper, we present an Unsupervised Feature Learning approach to Feature Description (UFL-FD), which is a 2D-feature description algorithm in feature space. Previous approaches mainly pay attention to multiscale space building such as Gaussian scale space (SIFT and SURF) and nonlinear scale space (KAZE).However, the building of those spaces is time-consuming and computation-expensive. In contrast, we describe the 2D-features in a feature space that is built by the unsupervised feature learning algorithm. In this way, the feature space is pre-built automatically form image itself to ensure the descriptor with accuracy and distinctiveness.Furthermore, in order to get objective appraisal of the UFL-FD algorithm, we present an extensive evaluation on benchmark datasets, which turns out that UFL-FD algorithm outperforms the SIFT, SURF and KAZE feature description algorithm in both accuracy and computation expense, yielding the algorithm that is of high quality.