Skip to Main Content
 

Global Search Box

 
 
 
 

Files

File List


18433.pdf (4.21 MB)

ETD Abstract Container

Abstract Header

Compact Image Signatures for Efficient Retrieval from Large GIS Raster Collections

Goparaju, Tejaswi
http://rave.ohiolink.edu/etdc/view?acc_num=ucin1447690962

Abstract Details

2015, MS, University of Cincinnati, Engineering and Applied Science: Computer Science.
The turn of the century brought with itself a significant increase in the volume of image data. We are familiar with large amounts textual data but with increased processing capabilities and new software techniques, we are able to generate and work with big image and terrestrial data. Querying over such data becomes a vital aspect when trying to identify patterns or similarities. Algorithms designed to aid processing of such huge amounts of data will be of immense importance to many fields such as medical, scientific, geographical domains. There are two commonly used methods for retrieving similar images based on a query image. First method is called Keywords or Tags based Image Retrieval. In this method, images similar to the query image are retrieved based on the keywords found in images’ metadata. Such a retrieval isn’t robust enough. To overcome this, Content Based Image Retrieval techniques have been developed. In these techniques, the underlying algorithms analyze the features of an image and match them for search and retrieval. This methodology is much more effective as the images are compared on basis of information held within them rather than their metadata. Our challenge is the design of a content based retrieval system for a dataset that consists of a large number of rasters of terrains. Each cell (similar to pixel in an image) has twenty different possible values denoted by terrain code values ranging between 11 and 95 (both inclusive). Our objective is to generate a compact signature for each image, representing its terrain characteristics, which can be used to compare with signatures of other images to identify and efficiently retrieve similar images. Such work has been done before but has not been applied to large GIS terrain datasets. When it comes to big datasets, having a compact signature is very important. We have designed four different types of raster signatures using various information theoretic methods and in this thesis we compare their performance. We use a retrieval algorithm designed on the lines of the Balanced Iterative Reducing and Clustering using Hierarchies (BIRCH) Algorithm for constructing the similarity tree. The goal of this research is to construct an efficient signature which describes the image and then use it to efficiently retrieve images/rasters with similar terrains. We show in this thesis that we have designed very effective signatures and retrieval algorithms. We use different metrics to generate the signatures for the Images such as the Raw Entropy, properties of Pixel Value Co-Occurrence Matrix such as entropy and energy and the Co-occurrence Matrix histogram distribution. We also show that despite being marginally behind the Co-Occurrence Matrix Probability Distribution signature in terms of retrieval performance, our designed compact signatures perform much faster and take significant less memory space for storage.
Raj Bhatnagar, Ph.D. (Committee Chair)
Nan Niu, Ph.D. (Committee Member)
Tomasz Stepinski, Ph.D. (Committee Member)
102 p.
Computer Science
Image retrieval; Compact Signatures

Recommended Citations

Citations

  • Goparaju, T. (2015). Compact Image Signatures for Efficient Retrieval from Large GIS Raster Collections [Master's thesis, University of Cincinnati]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=ucin1447690962

    APA Style (7th edition)

  • Goparaju, Tejaswi. Compact Image Signatures for Efficient Retrieval from Large GIS Raster Collections. 2015. University of Cincinnati, Master's thesis. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=ucin1447690962.

    MLA Style (8th edition)

  • Goparaju, Tejaswi. "Compact Image Signatures for Efficient Retrieval from Large GIS Raster Collections." Master's thesis, University of Cincinnati, 2015. http://rave.ohiolink.edu/etdc/view?acc_num=ucin1447690962

    Chicago Manual of Style (17th edition)

ucin1447690962
265
© 2015, all rights reserved.
This open access ETD is published by University of Cincinnati and OhioLINK.