Unsupervised Change Detection in Landsat Images with Atmospheric Artifacts: A Fuzzy Multiobjective Approach

Figure 8

Evaluation of the various methods on semisynthetic data set with various noise levels. (a) Synthetic data set with the existence of haze and thin clouds; (b) Lake Milh data set shown in Figure 3(b); (c) ground truth change mask. In (d), the image, shown in (a), is corrupted with different levels of zero mean Gaussian noise (from second row to fourth row, PSNR = 40, 20, and 10 dB, respectively). (e)–(i) shows the final change detection masks generated by DT-CWT, PCA- means, ERGAS, GA-PSO, and proposed method, respectively.

(a)

(b)

(c) Ground truth

(d) Corrupted image

(e) DT-CWT

(f) PCA- means

(g) ERGAS

(h) PSO-GA

(i) Proposed