Suppressing False Alarm in VideoSAR viaGradient-Weighted Edge Information
Round 1
Reviewer 1 Report
In this letter, a method for suppressing false alarm in VideoSAR via gradient-weighted edge information is proposed. The experimental results show the effectiveness. However, some issues need to be clearly clarified. My comments are:
In the abstract, the authors claim that this method could remove the false alarms. I would suggest to use the word “reduce” instead of “remove”. Page 3, the authors claims that the above 2 algorithms to form a novel difference-based shadow detection algorithm, and detailed information can be referred in [9]. If [9] is not published by the authors, I suggest not saying it a “novel” one. Moreover, what’s the relation with the proposed method? Is it a pre-detection module? Please explain it more clearly. It seems that the proposed method is applied on the real images. For the output of SAR systems, the complex image is always available. Is it also work for the complex images? Please explain it more. Please add references for the comparison methods.Author Response
Point 1: In the abstract, the authors claim that this method could remove the false alarms. I would suggest to use the word “reduce” instead of “remove”.
Response 1: I regard your expression 'reduce' as a better one, which means reducing the amount of false alarm, not remove them all. Thank you for your suggestion.
Point 2: Page 3, the authors claims that the above 2 algorithms to form a novel difference-based shadow detection algorithm, and detailed information can be referred in [9]. If [9] is not published by the authors, I suggest not saying it a “novel” one.
Response 2: To be honest, this 'novel' difference-based algorithm is proposed by myself, but the relevant paper has not been submitted yet. So in this paper I have to refer to a similar paper ([9]).
Point 3: Moreover, what’s the relation with the proposed method? Is it a pre-detection module? Please explain it more clearly.
Response 3: The mentioned difference-based algorithms are the pre-detection module, since the proposed false alarm reducing algorithm is operated after the pre-detection (several difference operations). And I will explain this more clearly in the next version manuscript.
Point 4: It seems that the proposed method is applied on the real images. For the output of SAR systems, the complex image is always available. Is it also work for the complex images? Please explain it more.
Response 4: VideoSAR is some kind a new thing in radar imaging. The real footage used in this paper is the only public real data all over the world (or freely available for me) published by Sandia National Laboratories of America. And Sandia didn't publish its complex data (only an mp4 file). So our research has to pause here before any complex VideoSAR data available for free. If any, I think that the complex data will definitely improve the detection effects.
Point 5: Please add references for the comparison methods.
Response 5: May I explain for my comparison experiment in the last Figure. Our proposed method for reducing false alarm is not conflict with other method mentioned in the Introduction section, they own a complementary relation. What I would like to point out is that our proposed method has different idea compared with other methods (consider the spatial distribution of false alarm). In terms of the comparison experiment, I want to compare the effect between unweighted edge detection and gradient-weighted edge detection, not the gradient-weighted edge detection method and other method.
Reviewer 2 Report
This letter proposes a Video SAR false alarm reduction method based on gradient-weighted edge information by utilizing the gradient difference between the target shadow edge and other edge regions in the image.
The idea is quite interesting, but there is a lot of confusions in equations that need clarifications:
Section 3.1
Equation 2 : notation \Delta a_s , figure 2 \delta a_1
Equation 3 : \Delta r_s ; \Delta r_1 (figure 1)
same for index m and 2
section 4 : equations (14) and (15)
what is function f
the two equations are suggesting the derivatives along x an y are the equal ( the right term is the same for the two equations)
line 136 : …the gradient of edge M(x,y) … this phrase is not clear : the gradient is (nabla f )
equation (16) : T_higth or T_i ?
equation (18) : what is N ?
Author Response
Point 1: Section 3.1, Equation 2 : notation \Delta a_s , figure 2 \delta a_1, Equation 3 : \Delta r_s ; \Delta r_1 (figure 1) . Same for index m and 2
Response 1: These are some writing errors and have been corrected.
Point 2: section 4 : equations (14) and (15), what is function f ? The two equations are suggesting the derivatives along x an y are the equal ( the right term is the same for the two equations)
Response 2: function f represents the image pixels. A figure has been added to give some detailed information.
Point 3: line 136 : …the gradient of edge M(x,y) … this phrase is not clear : the gradient is (nabla f )
Response 3: \nabla f has been deleted to make it more explicit.
Point 4: equation (16) : T_higth or T_i ?
Response 4: In equation (16) we obtain the value of T_high (corresponding to equation (14-15)). And the computing method of T_i is the same as equation (16). I have add some explanations in the latest version.
Point 5: equation (18) : what is N ?
Response 5: N is the total number of edge detection.
Reviewer 3 Report
The paper proposed a edge-based method to detect moving object more accurately. Overall the paper is well written and provide many interesting insights on SAR based object moving detection. These are some minor concerns that shall be addressed.
Are there any reference for Figure 1? And the reason why this causes false alarm to the system shall be described in there as well. Comparison to other state of the art methods (described in Introduction section) shall be performed. In addition to graphical figure, statisitical results shall also be presented to ease the comparison.Author Response
Point 1: Are there any reference for Figure 1? And the reason why this causes false alarm to the system shall be described in there as well.
Response 1: Figure 1 is summarized by myself with no reference. It is based on the experiment phenomena of real data. The reason why it is described here is that I want to give an intuitive concept of false alarm in VideoSAR. About how they form, where they most likely to emerge, etc. Furthermore, I also need to introduce the problem we prepare to solve in this paper, false alarm in edge regions.
Point 2: Comparison to other state of the art methods (described in Introduction section) shall be performed.
Response 2: Methods mentioned in Introduction section are not conflict with the proposed method. In this paper we propose a false alarm reduction method via edge information and it can be regarded as a supplementary method of other method like morphological filtering. If we consider the effects of these two method, respectively, morphological filtering performs better than our proposed method. But since the morphological filtering is kind of an indispensable procedure in difference-based detection algorithm. We didn’t consider the consequence of its absence. So in the latest version of manuscript, we add a table to list out the comparison between median filtering and our proposed method (without).
Point 3: In addition to graphical figure, statisitical results shall also be presented to ease the comparison.
Response 3: Statistical results are added in the latest version.
Round 2
Reviewer 1 Report
The quality of the revised manuscript is improved. However, several issues are not well clarified. I still have some concerns:
In frame number 70 and 630, the detection rate falls down for over 15%. Could you please explain more why this phenomenon happen? Or for more general case, which factor will lead to this? The proposed method is real—image based processing. I understand that real measured complex VideoSAR data is not easy to get. But I still suggest at least some words should be mentioned in the conclusion. For example, what’s the difference for SAR data, optical imagery, or other sensors as mentioned by the author. Because the main causes of false alarm for different sensors are different. The author claim that the proposed method has different idea compared with other methods (consider the spatial distribution of false alarm). Please clarify this issue in the context to avoid confusion. In the abstract, could you provide a quantitative descriptions of the merits of the proposed method? So the novelty is highlighted. Page 2, line 60, references should be provided for median average, mean average, and mixture gauss. Page 3, line 85, citation should be provided to Eq.(1). Page 175, citation should be made for the adaptive histogram equalization. Page 11. Citations should be provided for the comparison methods listed in Table I.Author Response
Point 1 : In frame number 70 and 630, the detection rate falls down for over 15%. Could you please explain more why this phenomenon happen? Or for more general case, which factor will lead to this?
Response 1 : As you suggested, we have add some explanations on this phenomenon in the relevant paragraph. It has been verified that the proposed method could reduce the amount of false alarms and do no harm to detection rate. But as VideoSAR is a complicated video data, the difference between real target and other false alarms are not distinct in all cases. Normally the reduction of false alarm would definitely be followed with the reduction of detection rate, since those method did not distinguish false alarm and real target. Our proposed method tries to distinguish these two objects via the edge information. Its effects have been tested by most frames in the video. But in frame number 70 and 630, its effects are influenced by a close-to-edge target.
Point 2 : The proposed method is real—image based processing. I understand that real measured complex VideoSAR data is not easy to get. But I still suggest at least some words should be mentioned in the conclusion. For example, what’s the difference for SAR data, optical imagery, or other sensors as mentioned by the author. Because the main causes of false alarm for different sensors are different.
Response 2 : As you suggested, we have revised some expressions in the conclusion and add some explanations to make it more clear.
Point 3 : The author claim that the proposed method has different idea compared with other methods (consider the spatial distribution of false alarm). Please clarify this issue in the context to avoid confusion.
Response 3 : Some explanations has been added to respond to your suggestion.
Point 4 : In the abstract, could you provide a quantitative descriptions of the merits of the proposed method? So the novelty is highlighted.
Response 4 : We sincerely accept your advice.
Point 5 : Page 2, line 60, references should be provided for median average, mean average, and mixture gauss.
Response 5 : Your valuable suggestion has been accepted.
Point 6 : Page 3, line 85, citation should be provided to Eq.(1).
Response 6 : Citation for Eq.(1) has been added.
Point 7 : Page 175, citation should be made for the adaptive histogram equalization.
Response 7 : Citation for the adaptive histogram equalization has been added.
Point 8 : Page 11. Citations should be provided for the comparison methods listed in Table I.
Response 8 : Citations have been added.
Reviewer 2 Report
The authors performed my suggestions, I have no further comments
Author Response
Thank you for your careful review. Wish you a nice weekend!