Robust Subspace Clustering with Block Diagonal Representation for Noisy Image Datasets

Round 1

Reviewer 1 Report

This paper presents a robust block diagonal representation approach (OBDR) for subspace clustering. By using L2,1 norm and F-norm, it can reduce the influence of noises on the construction of coefficient matrices and recover the spatial structure distribution of data. Experimental results show that the proposed algorithm generally outperforms other algorithms. Overall, the paper is well-organized. 

Some issues with the current writing include: 

In Section 4.3 Experimental Results, detailed analysis and explanation should be presented since OBDR does not consistently outperform other algorithms although it is generally effective. For example, as the authors pointed out, in Table 2, the results of D1 with 4 subjects shows that compared with OBDR, SBDR's performance does not vary much with the increasing of the noise levels, even though its performance dropped a lot from clean data to noise level 1%. SBDR performs even better than OBDR at noise level 5%. Similar things happen with some other table entries (e.g., results of D1 with 4 or 6 subjects w.r.t. SBDR on NMI in Table 3). What are the potential reasons causing this difference? Is it related to the algorithms' noise handling or is it dataset specific (i.e., only to D1)? If it is dataset specific, why and how do you ensure the performance of OBDR on other new datasets? What about the results if increasing the noise level, say 10%? Will that hurt the performance of OBDR more than other algorithms (e.g., SBDR)? Extra experiments should be conducted before drawing conclusions. 

Suggestion: lines 86-91, using a notation table instead of a paragraph to present the symbols and descriptions. 

From line 246 to line 248, to avoid the unnecessary plagiarism concerns, be careful while use words similar to the references (i.e., reference [11]); if tools like TurnItIn are available, please do similarity check of the manuscript by comparing the writing with other published articles. Also, the description of the map() function of ACC is inaccurate, please double check it (if needed, refer to that in reference [11]). 

Many of the references cited are a little old. More recent publications/results should be introduced and compared. 

Minor writing issues exist throughout the paper, for example: 

(1) line 16, insert "the" before "spectral", 
(2) lines 48-51, break the long sentence into two or three short ones. 
(3) line 283, insert "the" before "p-values"
...

Proofreading is needed before resubmission. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Article is well articulated and experimental design and results are well explained, how ever article requires improvement in presenting the comparing results with other existing models.

.Literature survey requires improvement with discussion on existing approaches with their challenges.

language polishing is required to improve article readability.

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Enjoyable read. Just a few comments regarding the conclusion. Would it be possible to spell out full abbreviations (i.e., OBDR) and also include the key findings from the study (the conclusions seem a bit brief!).

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report

Please find the attached file. 

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

I appreciate the authors' efforts to address the concerns in the first round of review. The new draft does show improvement in the writing, but there are still some issues to be fixed:

The resolution of Figure 1 is very low. 

As for the evaluation metrics of ACC and NMI (line 295), I suggest that the authors keep the definition details as they did in the first version, even though they need to slightly revise the words. Otherwise, critical information is missing from the paper. 

Line 177, Equation (7) and its number are not shown in the same line. The same problem happens with Equations 12, 17, 18 (i.e., lines 194, 210 and 216). Also, there are grid line issues with Tables 3 and 4 (extra lines at the end of page 10 and beginning of page 11, and at the end of page 11 and beginning of page 12). I am not sure if the paper was written in LaTex. If not, the authors may consider following the LaTex template to re-write it in order to fix the problem and improve the presentation quality, especially for equations, tables, and figures.

Lines 321, 331, 335, 336, 341, 344 and 370, "in the terms of" should be "in terms of". Also, the authors may consider substituting some of the "in terms of" since it was used many times within the same paragraph (line 319 to line 345). For example, lines 341 to 342 could be changed to "On the datasets D2 and D3, the OBDR significantly outperforms other algorithms with regard to ACC and NMI."

Lines 373 and 413, "in the view of statistics", change to "with regard to the evaluation metrics"

Lines 412-413, "the performance of OBDR has advantages over the compared algorithms", awkward! How can "the performance" have advantages over the compared algorithms? should be "the OBDR has advantages over the compared algorithms" or "the OBDR outperforms other algorithms". 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report

Please see the attached file. 

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.pdf