Graduate Thesis Or Dissertation

 

Evaluation functions in aerial image segmentation Public Deposited

Downloadable Content

Download PDF
https://ir.library.oregonstate.edu/concern/graduate_thesis_or_dissertations/kp78gk84r

Descriptions

Attribute NameValues
Creator
Abstract
  • Remote sensing is the most practical way to acquire large amounts of land cover data for monitoring and understanding environmental change, so it is important to be able to map land cover from imagery. Maps defining land cover patches as polygons rather than pixels greatly improve processing efficiency in models and are often more relevant to the scale of biophysical processes. Defining boundaries around homogeneous patches of an image is called segmentation. Assigning a cover type label to a pixel or polygon is called classification. Accuracy of maps generated from images is generally specified in terms of classification accuracy with no specification of spatial distribution of error. We demonstrate need for spatial accuracy assessment by showing that even small classification error rates can cause large changes in a standard landscape statistic, average patch compaction (APC). Furthermore, the common practice of reducing classification error by forcing a minimum mapping unit size can increase error in APC. We also investigate the importance of segmentation evaluation functions as the basis for scientific decisions, comparing algorithms, and driving optimizers. We divide these functions into reference-based (scores are derived through comparison to a "correct" reference segmentation) and heuristic (scores are based on attributes of the image without knowledge of a reference). We quantitatively measure performance of several evaluation functions embedded in a region merging algorithm on ecological images where hand segmentations are available. We explain their poor performance by showing that deterministic, irreversible segmentation algorithms like region merging can only represent n of 2[superscript n] possible segmentations for an initial segmentation containing n boundaries, suggesting that evaluation functions must be learned. We demonstrate a reference-based evaluation function as a target for learning. It achieves near optimal performance inside a region merging algorithm, showing that poor evaluation functions, not greedy algorithms, limit performance in region merging. Finally, we describe a prototype segmentation editing tool we built to simplify generating hand segmentations. This work is of interest to landscape ecologists computing metrics from maps derived from images, to remote sensing scientists evaluating their efforts in generating those maps, and to developers of segmentation algorithms and evaluation functions.
License
Resource Type
Date Available
Date Issued
Degree Level
Degree Name
Degree Field
Degree Grantor
Commencement Year
Advisor
Committee Member
Academic Affiliation
Non-Academic Affiliation
Subject
Rights Statement
Publisher
Peer Reviewed
Language
Digitization Specifications
  • File scanned at 300 ppi (Monochrome, 256 Grayscale) using Capture Perfect 3.0.82 on a Canon DR-9080C in PDF format. CVista PdfCompressor 4.0 was used for pdf compression and textual OCR.
Replaces

Relationships

Parents:

This work has no parents.

In Collection:

Items