Natural scene category recognition based on multiple channels of PHOW

This paper presents a method for recognizing scene categories based on multiple channels of Pyramid Histogram Of Words (PHOW). The main difference among different channels lies in what kind of feature detector/descriptor pair is employed in the framework of Bag-of-Words (BoW) models. This technique works by obtaining the confidence scores of a test image belonging to each possible category based on different information cues and combining those intermediate scores to determine the label of the test image. In order to make use of multiple cues provided by different channels of PHOW, we propose a novel fusion rule: weighted sum-max, which outperforms two other popular rules (max-max and sum-max) on several benchmark scene datasets.