Correlation-based variational change detection

---

Tezin Türü: Yüksek Lisans

Tezin Yürütüldüğü Kurum: Orta Doğu Teknik Üniversitesi, Mühendislik Fakültesi, Bilgisayar Mühendisliği Bölümü, Türkiye

Tezin Onay Tarihi: 2018

Öğrenci: GİZEM AKTAŞ

Danışman: FATOŞ TUNAY YARMAN VURAL

Özet:

Change detection is an important research topic for observing the earth but there are various challenges to obtain an accurate change map such as image noise, subtle differences and image acquisition alteration. Various studies handled these problems as separate steps. In the first step, images are registered, then the image noise are eliminated. After images are normalized to eliminate image acquisition differences, actual change detection routine can be applied as a final step. However, this step-by-step approach leads to accumulation of errors in each step which leads to decrease in accuracy of final change detection map. Step-by-step approach is widely used because these problems are interpenetrated to each other and researchers use this to divide problem into sub-problems. In this study, a correlation based variational change detection (CVCD) method for elevation models is proposed. In essence, CVCD aims to produce smooth change maps while preserving the details of the terrain by minimizing a variational cost function. In this variational cost function a novel correlationbased data fidelity term is used with an L1-norm regularization term which imposes smoothness on obtained change map. In addition, L1-norm TV regularization term is preferred because it can preserve details such as point-changes, edges and corners of changes. In order to minimize the proposed cost function using simple approximations in an iterative manner, a simple and efficient algorithm is suggested. Quantitative experiments on synthetic noisy data show that CVCD can provide a detection rate of 95% while staying in the low false alarm regime, i.e. less than 10^-2. Also, qualitative experiments on real-world data show the success of the CVCD for the changes with different characteristics.