Abstract
In recent years, volunteered-geographic-information (VGI) image data have served as a data source for various geographic applications, attracting researchers to assess the quality of these images. However, these applications and quality assessments are generally focused on images associated with geolocation through textual annotations, which is only part of valid images to them. In this paper, we explore the distribution pattern for most relevant VGI images of specific landmarks to extend the current quality analysis, and to provide guidance for improving the data-retrieval process of geographic applications. Distribution is explored in terms of two aspects, namely, semantic distribution and spatial distribution. In this paper, the term semantic distribution is used to describe the matching of building-image tags and content with each other. There are three kinds of images (semantic-relevant and content-relevant, semantic-relevant but content-irrelevant, and semantic-irrelevant but content-relevant). Spatial distribution shows how relevant images are distributed around a landmark. The process of this work can be divided into three parts: data filtering, retrieval of relevant landmark images, and distribution analysis. For semantic distribution, statistical results show that an average of 60% of images tagged with the building’s name actually represents the building, while 69% of images depicting the building are not annotated with the building’s name. There was also an observation that for most landmarks, 97% of relevant building images were located within 300 m around the building in terms of spatial distribution.
Highlights
Volunteered-geographic-information (VGI) images are data associated with a specific geographic location through visual content or metadata such as coordinates and textual descriptions [1]
This information has been used in various geographic applications in recent years, such as place-semantic-information extraction [2,3,4], scene summarization [5,6,7], and 3D reconstruction [8,9,10,11,12,13,14]
The datasets used in the present study were downloaded from the Flickr APIsbausesdedonitnheitrhleocaptiroensesn. tWestruetdryievewdearell pduobwlicnllyoaadvaedilabflreoamnd tgheeotaFglgicekdrFlAickPrI i(mhtatpgse:s//owf wLown.dfloicnkfrr.coomm2/s0e1r6v, iacnesd/aapcqi/u) ibreadsemd oorne tthhaenir4l8o9c,a0t0io0nims. aWgees rientrtoietvale.dAafltlerptuhbelidcalytaa-fivlatielraibnlge and geotagged Flickr images of London from 2016, and acquired more than 489,000 images in total
Summary
Volunteered-geographic-information (VGI) images are data associated with a specific geographic location through visual content or metadata such as coordinates and textual descriptions (e.g., location name) [1]. Some content-related images are normally missing due to imprecise descriptions in the annotated tags Another way to retrieve VGI images is to search images close to a specific geolocation, but how much of the area surrounding the place should be searched remains an open question. Semantic-distribution patterns show the accuracy of textual annotations (i.e., tags) by providing the proportion of images that were both tagged with the name of the given landmark and that visually represents the landmark. This should be an important value to set initial search regions for research communities when searching for images of specific buildings.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
