A Novel Horizon Picking Method on Sub-Bottom Profiler Sonar Images
Round 1
Reviewer 1 Report
---
SBP Images:
It would be usefull to illustrate building the SBP Images (example image before and after).
---
Multi-scale filter:
the authors say, that the -Hessian matrix "can capture" the intensity structure of the local neighborhood-
Later on it is a simple convolution: -When structures of two signals are similar, the convolution between these signals is high-
These statement is not a novel finding. It is furthermore not really clear in which way the filter is fully applied. For example the statement -RB reaches a maximum- is confusing. The question is here along which function/variable ?
The statement "the second-order derivative of a Gaussian kernel, the interval [-s, s], corresponds to the interface area" is confusing. In case the filter is calculated for different scales, it should be better explained ?
Interval [-s, s] ? Is this the parameter interval or spatial interval of the multiscale filter ?
-
Concerning the given references it seems to be strange that only conference papers are given which also can't provide more details. One is within the same application [14] and the other is already from 1998 in the field of medical image processing.
The application of such a Multi-scale filter to such problems is actually intersting but I think the author should use the chance the explain the application of this filter in more detail. For example providing also more interims results when processing data to detect the lines.
On the other hand I think figure 2 is not necessary. What does it help to show a gaussian function ?
---
The Verticality Descriptor:
This section is difficult to understand. It is not really clear it which way the IM() data is processed.
Remarks/questions to this sections are:
- "If both l1 and l2 are small, the analyzed region" what is a region ?
- figure 4a not really necessary, while 4b deserves some more explanations.
- (17)/(19) double definition - necessary ?
- for what are the derivatives (12) exactly good for ?
---
Experiments/Results:
- Figure 6 - what exactly is different in the red boxes compared to neighboring regions ?
-> Local phase has been used to highlight reflections when lacking full-waveform data [10], [13], but it is not robust enough and some pseudo structures may be highlighted too ?
It would be interesting to emphasize these problems from previous paper to new method.
- envelope data, full waveform data - differences/effects with respect to the proposed method would interesting to show and explain in more detail.
minor things:
- l187 -> ... it is necessary ...
- eqn. grouping is confusing (x...,y...)
- Frangi... better to cite always with the number [16]
- [lb, ub] corresponds to [0, 255]. ? the range 0 -> 255 ?
Author Response
Please see the attachment.
Author Response File: 
Author Response.doc
Reviewer 2 Report
Authors propose a method for automatic horizon picking from sub-bottom profiles. They proposed better multi-scale enhancement filtering algorithm. A vertical suppression weightening term based on the form of logistic function is applied. The topic is interesting especially that standard manual methods are time-consuming. The good point is that the proposed method has been verified experimentally with different equipment and in various test areas. I would like the article to be published after revisions.
Some comments:
1. Line 30: Sub-bottom profilers are designed to obtain the images of underwater subbottom
profiles (SBP) – authors already developed acronym SBP in line 15. You do not
have to described acronym twice.
2. Line 31: incorrect reference citation: and other tasks [1]-[4]. See the instructions for
authors.
3. Figure 1: develop the acronym PSO in the figure 1, please.
4. Line 106: There is no space between dot and letter A: 1.A
5. Line 109: There is no space between dot and word Adjacent: 2.Adjacent
6. Line 109: There is no space between ratio abd bracket: signal noise ratio(SNR)
7. Line 133: I am missing a reference in this sentence: Frangi used the geometric
characteristics derived from the Hessian matrix to construct descriptors suitable for
line-like structures.
8. Line 145: correct the spelling mistaje: By combining the two descriptors, a line-like
structure filter can be been built.
9. Line 187: probably “is” is missing: To overcome the problem, it necessary to improve
multi-scale vessel
10. Line 246: I would suggest to present equations for key points “a” and “d” in two
separate lines
11. Line 247: missing dot behind digit 3. 2.4.3 Determination of Key Points and
Parameters
12. Line 284: This subsection 2.4.3. is too short to be as a separate one. Please try to
combine this subsection with another one. It is only one sentence and one equation.
13. Line 321: EdgeTech is written probably together EdgeTech not Edge Tech. Horizon
Picking on Edge Tech 3100P Data in ShenZhen
14. Line 324, 391, 392, : as above
15. I am missing at least one chart where the survey was conducted. Please provide one
more figure with all test areas.
16. Line 427: …Horizon Picking onParasound P70 Data inSouth China Sea… – missing
space.
17. Line 429: please provide the company of Parasound P70 (in bracket).
18. Line 454: ……….sub- bottom…………... Useless space
19. Line 495-496: Thus, based on this characteristic, the proposed method achieved
excellent performance. The proposed method can be applied to both of…………. –
improve the sentence, please, word repetition mistake.
Author Response
Please see the attachment.
Author Response File: Author Response.doc
Round 2
Reviewer 1 Report
Dear Authors,
the criticised points are better explained. But, if possible, I still would appreciate additional references to the multiscale filter.
Nevertheless, I would accept the paper in the present form.
best regards
