Bag of Visual Words Model with Deep Spatial Features for Geographical Scene Classification.

Jiangfan Feng,Lin Wu,Yuanyuan Liu

doi:10.1155/2017/5169675

Abstract

With the popular use of geotagging images, more and more research efforts have been placed on geographical scene classification. In geographical scene classification, valid spatial feature selection can significantly boost the final performance. Bag of visual words (BoVW) can do well in selecting feature in geographical scene classification; nevertheless, it works effectively only if the provided feature extractor is well-matched. In this paper, we use convolutional neural networks (CNNs) for optimizing proposed feature extractor, so that it can learn more suitable visual vocabularies from the geotagging images. Our approach achieves better performance than BoVW as a tool for geographical scene classification, respectively, in three datasets which contain a variety of scene categories.

Highlights

Geographical scene classification can effectively facilitate the environmental management, indoor and outdoor mapping, and other analysis in dealing with spatial data
The traditional bag of visual words (BoVW) model tends to require the identification of the spatial features of each pixel with a small number of training samples
Since each image contains a set of low-dimensional feature vectors, which has similar structure as dense SIFT, we propose to encode these feature vectors into a single feature vector using standard BoVW encoding

Summary

Introduction

Geographical scene classification can effectively facilitate the environmental management, indoor and outdoor mapping, and other analysis in dealing with spatial data. Spatial features have been proven to be useful in improving the representation of the image and increasing classification accuracies. A widely used method named bag of visual words (BoVW) finds the collection of local spatial features in the images and combining appearance and spatial information of images. The traditional BoVW model tends to require the identification of the spatial features of each pixel with a small number of training samples. Despite the fact that their advantages have been proven in small data sets, their performance is variable and less satisfactory while dealing with large datasets due to the complexity and diversity of landscape and land cover pattern. It is a challenge to obtain higher interpretation accuracies when dealing with increased geographical imagery

Objectives

Results

Conclusion