A Hybrid Information Reconstruction Algorithm for Multitemporal Landsat Image

論文翻譯標題: 混合式資訊重建演算法應用於多時期Landsat衛星影像修復
  • 陳 誌彬

學生論文: Master's Thesis


The key to information reconstruction of cloud-contaminated satellite images is to recover missing data by utilizing temporal and contextual information while maintaining radiometric accuracy and consistency Most previous studies achieved this objective by using patch-based information cloning or pixel-based contextual prediction Patch-based methods that utilize temporal correlation of multitemporal images have the advantage of radiometric consistency whereas pixel-based methods that use spatial contextual information can achieve radiometric accuracy A hybrid method that integrates patch-based cloning with pixel-based prediction is proposed to provide a radiometric accurate and consistent reconstruction In the proposed method a small set of cloud-contaminated pixels with high-confidence filling results is determined on the basis of the fact that same-class pixels have similar spectral characteristics and exhibit similar temporal changes between dates These pixels which are called fixed points are used to optimize patch-based radiometric cloning Radiometric patch cloning is mathematically formulated as a Poisson equation and solved by using an optimization process Several cloud-free and high-similarity patches are optimally cloned to a corresponding cloud-contaminated region under constraints from fixed pixels Cloning optimization can lead to radiometric consistent results Fixed-point constraints can improve radiometric accuracy by reducing error propagation in radiometric cloning In experiments simulated images and actual image sequences acquired by Landsat Enhanced Thematic Mapper Plus sensor are used to assess the performance of the proposed hybrid method Experimental results indicate that our method can accurately recover the value of cloud-contaminated pixels and the reconstruction accuracy is improved in comparison with related methods
獎項日期2015 一月 30
監督員Chao-Hung Lin (Supervisor)