RPI-SURF: A Feature Descriptor for Bubble Velocity Measurement in Froth Flotation With Relative Position Information

Abstract

The local feature-based method is the most popular choice for froth velocity measurement. However, due to similar patterns or objects in the froth image pair, current local feature-based methods only consider local intensity character (LIC) and cannot extract the matches well. Therefore, we propose a new feature descriptor named relative position information (RPI)-SURF that integrates the LIC and the RPI from the nearest reference point. First, we use the SURF descriptor with a distance constraint to extract the LIC. Next, we calculate the RPI by the direction and distance to the nearest reference point pair. After that, we integrate the LIC and RPI to demonstrate the new feature descriptor. Extensive experiments verify the effectiveness of the proposed feature descriptor. Compared with the SURF descriptor, the recall and $F_{1}$ -score of the proposed RPI-SURF descriptor have been obviously increased, and the number of detected accurate matches on real froth image pairs by the RPI-SURF descriptor has been increased by at least nearly 60%.