Performance Comparison of Matching Cost Functions for High-Quality Sea-Ice Surface Model Generation
Cited 0 time in
Cited 0 time in
- Performance Comparison of Matching Cost Functions for High-Quality Sea-Ice Surface Model Generation
- Other Titles
- 고품질 해빙표면모델 생성을 위한 정합비용함수의 성능 비교 분석
- Kim, Jae-In
- Other natural science
- Digital surface model; Sea Ice; aerial image; Matching cost
- Issue Date
- Kim, Jae-In, Kim, Hyun-cheol. 2018. "Performance Comparison of Matching Cost Functions for High-Quality Sea-Ice Surface Model Generation". Korean Journal of Remote Sensing, 34(6-2): 1251-1260.
- High-quality sea-ice surface models generated from aerial images can be used effectively as field data for developing satellite-based remote sensing methods but also as analysis data for understanding geometric variations of Arctic sea-ice. However, the lack of texture information on seaice surfaces can reduce the accuracy of image matching. In this paper, we analyze the performance of matching cost functions for homogeneous sea-ice surfaces as a part of high-quality sea-ice surface model generation. The matching cost functions include sum of squared differences (SSD), normalized crosscorrelation (NCC), and zero-mean normalized cross-correlation (ZNCC) in image domain and phase correlation (PC), orientation correlation (OC), and gradient correlation (GC) in frequency domain. In order to analyze the matching performance for texture changes clearly and objectively, a new evaluation methodology based on the principle of object-space matching technique was introduced. Experimental results showed that it is possible to secure reliability and accuracy of image matching only when optimal search windows are variably applied to each matching point in textureless regions such as sea-ice surfaces. Among the matching cost functions, NCC and ZNCC showed the best performance for texture changes.
- Files in This Item
- Can archive pre-print and post-print or publisher's version/PDF
Can archive post-print (ie final draft post-refereeing) or publisher's version/PDF
Can archive pre-print (ie pre-refereeing)
Archiving not formally supported
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.