Performance evaluation for feature extractors on street view images

摘要

This paper comprises a comprehensive evaluation of popular feature extractor algorithms used in the analysis of street view images. Feature extractors are considered to consist of two main parts, namely, feature detectors and feature descriptors. This study aims to evaluate state-of-the-art feature extractors under different experimental conditions. Hybrid algorithms generated from current feature detectors and descriptors are tested under the same experimental conditions. The expected outcome of the experimental procedures should facilitate the selection of distinctive and robust extractors for different viewing conditions.