Thick Cloud Removal in Multi-Temporal Remote Sensing Images via Frequency Spectrum-Modulated Tensor Completion

Round 1

Reviewer 1 Report

As described in Fig 1, each band is arranges as an mxnxt tensor and the authors claim on line 109 that each of these tensors is low rank. However, this requires each mxn spacial matrix in the tensors to be low rank and this is rarely true for images in general because the spatial signal typically does have have a high degree of correlation.     Presumbly the set defined by omega is previously identified using a cloud detection model, but this is not discussed.     T_omega is not defined, and apparently is a low rank tensor model of the data as suggested by eqn 2 on line 115.     Please explain the objective given in eqn 1 at line 112. It is unclear why the rank of X (the measured spatial temporal matrix at a single band) is to be minimized and it is not clear why the min is take wrt the measurements and not the model parameters. Same commen for eqn 2.     It is not clear why the trace norm (total sum of squares of the measurements) is added to the mimimization criterion. It is a constant for each band.

Author Response

Thank you for your careful evaluation of this manuscript. We have tried our best to improve the manuscript according to your valuable comments. Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

This paper presents a frequency-spectrum modulated tensor completion method for multi-temporal remote sensing images. FMTC rearranges the original four-dimensional tensor to establish a series of spatio-temporal tensors with low-rank properties, where the Fourier transform is introduced in the temporal dimension. The paper is reasonably we written and well explained. 

However, there are few observations to improve the paper:

1.     FMTC is used but not defined anywhere  

2.  Clearly discuss the novelty and contribution of the paper at the end of introduction section

3.     Clearly mention the limitation of existing work and motivation of the proposed work

4.     Briefly discuss the results of proposed technique in abstract and conclusion

Author Response

Thank you for your careful evaluation of this manuscript. We have tried our best to improve the manuscript according to your valuable comments. Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

The authors propose a frequency-spectrum modulated tensor completion method to remove clouds. The method is well presented, and experimental results show its advantage.  To make this work convincing, the following comments should be addressed.

(1) The novelty is not clear and limited. Why does FMTC outperform conventional tensor completion?

(2) The experimental results are not convincing. For simulated data, the FMTC outperforms ST-tensor by a large margin. However, for real data, the recovery results of these two methods are similar. Please clarify this.

(3) For experimental results on simulated data, the recovery results are the same as the source images. The proposed method achieves amazing performance, which is not convincing. Especially for Figs. 6, 10, and 12, most information is missing, and the reconstructed results are also the same as the source image.  

(4) Fig. 13 is missing in this manuscript.

 

 

Author Response

Thank you for your careful evaluation of this manuscript. We have tried our best to improve the manuscript according to your valuable comments. Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 3 Report

The authors have addressed all my concerns, and this work is suggested to be accepted.