Talk:Computer vision/Archives/2015
![]() | This is an archive of past discussions. Do not edit the contents of this page. If you wish to start a new discussion or revive an old one, please do so on the current talk page. |
About the "Methods of Computer Vision" section
- First two posts are copied from the users talk pages --KYN (talk) 20:11, 12 August 2009 (UTC)
Hi, I'm sure there is something interesting to say about Boustrophedon cell decomposition in relation to computer vision, but I don't believe your recent edit in the last article was the right way. The CV article is an overview of this rather large and disparate area, it mentions breifly various application but not any of the methods that are used to solve the different tasks. Therefore, I believe your edit came out too tecnical to really fit the rest of the article.
Is it possible that, instead, you expand the Boustrophedon cell decomposition article, write an overview on the 3D reconstruction problem (where this specific technique appears to fit in) and then we can expand the 3D reconstruction part of the CV article and link it to that review? --KYN (talk) 19:24, 10 August 2009 (UTC)
- I see your point. My initial goal was to help remove the orphan tag on the BCD page as very little makes reference to it. You suggested instead to make a reference to BCD in the 3D reconstruction part of the CV article. I could agree with this. The only thing I can't do is provide an overview of the BCD on the BCD page! I created the page about a year ago when I first heard of the concept in my AI class, but truth be told, I am far from an expert on the particular method... I do feel strongly that in some way, related topics in general should be interlinked. Separating theory and application is wonderful, but it should be possible for a user to read information on one type of page and be able to arrive at the other type, in my opinion. ARF (talk) 14:38, 11 August 2009 (UTC)
You are right that related topics should be interlinked, and you can use the standard WP:wikilink for that purpose, it is not necessary to write text about a related topic in the article. My main concern is that if everyone writes a little something about their favorite computer vision methods in this article (and there are plenty of examples that could fit in here) it will no longer be a reasonable overview of the area that someone not already familiar with it can read and understand. Here's my suggestion:
- Remove the recent edit in this article (motivated above), possible reusing it in the overview I suggested in case you feel inclined to write it.
- Add the Boustrophedon cell decomposition article to Category:Computer vision. That way it becomes accessible to anyone looking for more detailed information about the CV area.
- Expand the Boustrophedon cell decomposition to give a better explanation on how it connects to the CV area. In that way it also becomes easier for other CV related articles to link directly to it, making it less of an orphan. --KYN (talk) 20:31, 12 August 2009 (UTC)
- Ok, I like your suggestions. I cannot do that much right now, but I have put it on my to-do list and will take care of this as soon as I can. Perhaps once I get out of work for the day! Thanks Kyn for your constructive dialog. ARF (talk) 17:23, 14 August 2009 (UTC)
BCD doesn't seem to be related at all to computer vision. It appears to be a path-planning method, so it would go there, if included on Wikipedia at all. But a quick look at Google Scholar citations suggests that it's probably not important enough yet to merit a Wikipedia entry. --209.90.141.207 (talk) 22:15, 3 September 2011 (UTC)
Definition in the lead
Right now the lead sort of uses a (synonym of) a word to define itself. (Computer Vision = "Machines that see") We really need to define "see". (e.g. automatically extracting the required information from the image, including decision making). Not that I'm ready to tackle this right now! :-) North8000 (talk) 15:08, 23 August 2010 (UTC)
- If you want to present to the not already informed reader what CV is about, "see" could be a reasonable way to start, even if I agree that it is both vague and possibly also incorrect. I tried to make the description more precise. --KYN (talk) 12:50, 26 August 2010 (UTC)
"See Also" List
As someone noted it is too long. But 2/3 of it is an excellent list. A difficult but worthwhile job would be to reduce it while leaving in the best stuff. North8000 (talk) 10:43, 1 December 2010 (UTC)
- Even better would be to turn some of those into cited sources. Dicklyon (talk) 17:31, 1 December 2010 (UTC)
Revised lead sentence by 86.....
It's sort of wrong (Machine Vision is not an exact synonym) but I did not revert I think overall it was an improvement. We should tweak it. North8000 (talk) 12:20, 10 June 2011 (UTC)
- But now you did revert it out. I think I'll put it back in; it is good to know that machine vision is another term for pretty much the same topic. Dicklyon (talk) 07:18, 4 September 2011 (UTC)
Merge Proposal with Article "Machine Vision"
The content stated in "Machine Vision" just refers to an application of computer vision (First sentence of the article "Machine vision is a branch of engineering that uses computer vision in the context of manufacturing"); they are essentially belonging to the same topic. It is desirable to merge these two articles to reduce redundancy and overlapping of information. Jacob-Dang 15:46, 22 July 2011 (UTC)
- Getting the Machine Vision article to rise from it's ruins is still a work in progress. The lead sentence there needs re-writing.
- While there is some overlap technologically, the common meanings and uses of these terms are very different. North8000 (talk) 15:58, 22 July 2011 (UTC)
- While I am not experienced at machine vision, after reading the article my impression is that much of the content is quite general and similar to computer vision, such as "Imaging" and "Image Processing" section. If we want to keep the article, we shall really focus on those difference.Jacob-Dang 16:45, 22 July 2011 (UTC)
- You are right. This is the true overview stuff on machine vision....very slow & hard to work in and source in the adverse environment there (see talk page months back). Partially because image processing (something which IS in common with computer vision) in the emblematic part of machine vision, but in reality is just 25% of what machine vision is about. Sincerely, North8000 (talk) 18:18, 22 July 2011 (UTC)
- Thank you North8000 for your explanation. Let's us wait for more people to participate in this discussion. If there is no sufficient support than after around 2 weeks I will remove this tag. Thank you. Jacob-Dang 06:52, 24 July 2011 (UTC)
- That sounds good. It is a good question to raise and get clarified/decided on. BTW, I had been planning to eventually revise that MV lead sentence and now have done it. North8000 (talk) 12:34, 24 July 2011 (UTC)
- It may help the discussion if someone can present an outline of the implementation of the proposed merge. --KYN (talk) 15:02, 24 July 2011 (UTC)
- I know I can't because I don't feel that the proposed merge is a good idea. North8000 (talk) 02:50, 28 July 2011 (UTC)
- Jacob-Dang, reinforcing my earlier comment, even though I don't think that the merge is a good idea, I think that the discussion should be fully played out...even if it takes a few months to gather additional comments/ have additional discussion. So I would not want to end the proposed merge discussion too quickly. Sincerely, North8000 (talk) 02:54, 28 July 2011 (UTC)North8000 (talk) 23:25, 3 September 2011 (UTC)
- It may help the discussion if someone can present an outline of the implementation of the proposed merge. --KYN (talk) 15:02, 24 July 2011 (UTC)
- That sounds good. It is a good question to raise and get clarified/decided on. BTW, I had been planning to eventually revise that MV lead sentence and now have done it. North8000 (talk) 12:34, 24 July 2011 (UTC)
- Thank you North8000 for your explanation. Let's us wait for more people to participate in this discussion. If there is no sufficient support than after around 2 weeks I will remove this tag. Thank you. Jacob-Dang 06:52, 24 July 2011 (UTC)
- You are right. This is the true overview stuff on machine vision....very slow & hard to work in and source in the adverse environment there (see talk page months back). Partially because image processing (something which IS in common with computer vision) in the emblematic part of machine vision, but in reality is just 25% of what machine vision is about. Sincerely, North8000 (talk) 18:18, 22 July 2011 (UTC)
- I've looked at sources on this issue, and I find two sorts that mention both "computer vision" and "machine vision": A few contrast these terms, and say the "machine" term is for more applied and industrial stuff, while "computer" is for more research and algorithms. The others just use the two terms interchangeably. There is clearly a huge overlap, and no clear consensus on the distinction, though in some communities there are usage differences as pointed out. On balance, I think a merge might be OK. Dicklyon (talk) 07:17, 4 September 2011 (UTC)
Places where the term machine vision was/is placed as a synonym for computer vision
I took out 3 instances where the term "machine vision" was used as a substitute/synonym for "computer vision". Dicklyon felt otherwise and reverted those. At least per the common (industrial) use of "machine vision" (which I am extensively familiar with) it would not be applicable. But for all I know, there may be other common meanings of the term "machine vision" within computer vision fields which I am not aware of. Posted for discussion. North8000 (talk) 11:00, 4 September 2011 (UTC)
Even though there is a few discussion here however it has been months after I started the disucssion. Also I think "North8000" is trustworthy for this subject. Jacob-Dang — Preceding unsigned comment added by JacobDang (talk • contribs) 12:11, 5 September 2011 (UTC)
3D material
Wrote this to the person who added the material: It's nice to have someone knowledgeable in that field add material. The challenge is that in it's current state it's sort of specialty sidebar comments, and on one particular method (2d image based) of 3d. Would you be interested in writing it to be more of an introduction and explanatory? Thanks. Sincerely, North8000 (talk) 15:46, 27 September 2011 (UTC)
- After editing that section I started to wander Wikipedia. There are other articles which probably should be merged or referenced in that section:
DMahalko (talk) 22:47, 2 October 2011 (UTC)
- Thanks. The scanning based 3D is the one I'm much more familiar with. I believe that the current article material is about 3D constructed from multiple images; I'm not knowledgeable that area and was hoping someone could tweak it. Sincerely, North8000 (talk) 23:55, 2 October 2011 (UTC)
- Another concern was expressed below about this section. And despite multiple requests over time (including two requests on the talk page of the person who put it in) )I have not been successful in my request for the inserter to edit this or even directly address the question. This has the look of a narrow section copied from a broader writing that actually explained the topic. As such, with actual explanatory wording on the topic completely missing, in this context it is of little or no value to readers, and also raises some copyvio concerns. Also the illustration really isn't related to the topic, it is an older picture representing a way of presenting the topography of Mt. St. Helens. Not a terrible choice, but the point is that no CV content would be lost by removal of the image. I reluctantly suggest that we delete this material. Sincerely, North8000 (talk) 17:25, 20 November 2011 (UTC) North8000 (talk) 13:07, 21 November 2011 (UTC)
- Despite multiple attempts, I've been unable to get DMahalko to take this further, and I am unable to do so. Currently it is random intelligent notes that add up to gibberish. Reluctantly, I am going to remove it. North8000 (talk) 13:04, 12 January 2012 (UTC)
Workspace for definition
- Answers.com: "The use of digital computer techniques to extract, characterize, and interpret information."North8000 (talk) 10:49, 29 September 2011 (UTC)
- The Free Dictionary/Computer Desktop Encyclopedia: "Using a computer to capture and analyze images"North8000 (talk) 10:49, 29 September 2011 (UTC)
- The Free Dictionary/The McGraw Hill dictionary:
- " The use of digital computer techniques to extract, characterize, and interpret information in visual images of a three-dimensional world."North8000 (talk) 10:49, 29 September 2011 (UTC)
- "The technology concerned with computational understanding and use of the information present in visual images"North8000 (talk) 10:49, 29 September 2011 (UTC)
- The Free Dictionary/FOLDOC-The Free On Line dictionary of computing: "A branch of artificial intelligence and image processing concerned with computer processing of images from the real world. Computer vision typically requires a combination of low level image processing to enhance the image quality (e.g. remove noise, increase contrast) and higher level pattern recognition and image understanding to recognise features present in the image."North8000 (talk) 10:49, 29 September 2011 (UTC)
- Dictionary.com: "a robot analogue of human vision in which information about the environment is received by one or more video cameras and processed by computer: used in navigation by robots, in the control of automated production lines, etc."North8000 (talk) 10:49, 29 September 2011 (UTC)
Conclusion: The current first sentence of the article is a pretty good definition. North8000 (talk) 10:49, 29 September 2011 (UTC)
About recent edits
There are been a large number of edits of the Computer Vision article over the last few weeks. I have reviewed these edits (including some older as well) and have some comments about some of them:
- The very first sentence in the lead is problematic. The word automated is ambiguous and have at least two meanings: (1) done by a machine/computer, (2) done without other interventions. I'd say (1) is OK but (2) not since CV in some cases is used with human intervention (e.g. choosing which targets to track, or which parameters to use in a specific method). Why not use a more precise description?
- I see the word "automatic" as both unnecessary if we instead use "computer based", and also incorrect since computer vision does not exclude human invervention. Therefore I want to remove "automatic" from the lead. --KYN (talk) 14:59, 11 November 2011 (UTC)
- On the latter point, you are making it sound as if saying the the processing side is automatic equates to saying human interaction is excluded from the entire process. It does not equate to asserting that. But IMHO the fact that the processing is automated is important to making the distinction. After all, somebody manually doing image processing using a computer (e.g. using Photoshop) is not computer vision, but the definition with "automated" removed would say that it is. Sincerely, North8000 (talk) 15:17, 11 November 2011 (UTC)
- I don't understand your arguments, can you be more precise about what is lost if we remove the word "automatic"? --KYN (talk) 15:37, 19 November 2011 (UTC)
- In formulating an answer I realized that my addition filled only one of the two key missing elements left by the deletion. I will add the other one and then provide a response here. North8000 (talk) 16:49, 20 November 2011 (UTC)
- A definition which has insufficient specificity to the extent that it includes things which are clearly not computer vision. Hence both the (automated) imaging process and the automated processing of the images are central to a usable definition. Without those, (i.e. per the recent change which deleted the lead sentence and substituted another one), any uses of computer processing of images would be defined as computer vision. For example, somebody simply using Photoshop on a stored photo would have been defined as "computer vision" which is clearly in error. North8000 (talk) 17:02, 20 November 2011 (UTC)
- In formulating an answer I realized that my addition filled only one of the two key missing elements left by the deletion. I will add the other one and then provide a response here. North8000 (talk) 16:49, 20 November 2011 (UTC)
- Ok, so if we agree on that computer vision is whatever processing can be done on images as long as it is not made in Photoshop, then automatic is not a word to make the distinction. It is perfectly possible to set up Photoshop (or similar programs) to do automatic processing of images, so using automatic does not lead away from Photoshop. The key issue, IMHO, is the end of the first sentence: ... to extract and interpret information. This is what you typically cannot do in Photoshop. Since that part of the sentence is already there, we don't need automatic since that in an unnecessary way may lead the reader, who has no previous information about what computer vision is, to the conclusion that computer vision must happen without any human interaction. This is clearly not the only interpretation of what automatic means, but since it cannot be used to distinguish computer vision from Photo-shopping I don't really see this point of having it there. (PS not my edits of the lead since we started this thread) --KYN (talk) 19:47, 22 November 2011 (UTC)
- Well to start with, I certainly agree that the extraction of information is central to the definition. But if there is no qualifier such as "automatic" or "real time". What do you think? If you don't agree then feel free to take it out or I'll take it out. While I think that taking it out makes the definition less useful, that particular item is less important than having a nice working atmosphere here. Sincerely, North8000 (talk) 21:35, 22 November 2011 (UTC)
- I see the word "automatic" as both unnecessary if we instead use "computer based", and also incorrect since computer vision does not exclude human invervention. Therefore I want to remove "automatic" from the lead. --KYN (talk) 14:59, 11 November 2011 (UTC)
- One way to resolve this issue, and also address the ever appearing discussion about lack of references is to consult the text books. I a few such text books, each of them tring to present a universal definition of what computer vision is about:
- (Sonka, Hvalac, Boyle) ... computer vision aims to duplicate effects of human vision by electronically perceiving and understanding an image
- (Morris) … computer vision involves extracting numerical or symbolic information from images
- (Shapiro & Stockman) The goal of computer vision is to make useful decisions about the real world based on images
- (Forsyth & Ponce) We see CV ... as an enterprise that uses statistical methods to disentangle data using model constructed with the aid of geometry, physics and learning theory
- (Jähne & Haussecker) computer vision is understood as the host of techniques to acquire, process, analyse, and understand complex higher-dimensional data from our environment for scientific and technical exploration
- Note that neither of them use the qualifiers automatic or real-time. Trying to produce a combo-statement about what computer vision is from all of these is not straight-forward, but I have replaced the first paragraph of the lead with a first draft of such a combo. --KYN (talk) 14:49, 29 November 2011 (UTC)
- Good research. As indicated previously, if you want to change it, that's fine with me. Sincerely, North8000 (talk) 19:04, 29 November 2011 (UTC)
- Didn't want to just jump in while you working on it, but I'd be happy to help with those ref issues. North8000 (talk) 22:03, 29 November 2011 (UTC)
- Good research. As indicated previously, if you want to change it, that's fine with me. Sincerely, North8000 (talk) 19:04, 29 November 2011 (UTC)
- One way to resolve this issue, and also address the ever appearing discussion about lack of references is to consult the text books. I a few such text books, each of them tring to present a universal definition of what computer vision is about:
- (Continuing with the first sentence) What exactly does processing of real world images mean? Is it inserted to exclude the possibility of processing synthetic images produced by a computer graphics system? If yes, what is it called when such images are processed, if "no" do we really need the "real world" specifier?
- (Continuing with the first sentence) Computer Vision does not exclude non-real time processing (what ever real-time means in this context). We may even make a distinction between methods in computer vision and their implementations. A specific method may or may not have real-time implementations. Also, real-time is a ambiguous word, it may mean being able to process images in video rate, or being able to produce a result in time for a system to solve a specific task, e.g., producing a diagnosis to a patient within less than a week. As an example, image retrieval from image databases is typically not done in real time (unless we accept that real time may take a few seconds for a really big database, or use very special type of hardware solutions) but tracking can typically be done in (video) real-time.
- There are now two places in the lead where it mentions computer vision applications: In the second paragraph and in the dotted list at the end. Why is this so?
- The lead now declares a distinction between computer vision and machine vision. As has been discussed both here and on the MV talk page, such a distinction may or may not exist, and is less or more clear, depending on personal views. This in not reflected in the lead. If we should have this type of declaration here, it should reflect this issue. Also, there seems to be research groups/companies that claim to do machine vision, e.g.,in the medical area, so it is not restricted to industrial/robotic applications.
- Overall, I don't see a clear line in the edits of the lead. What was wrong with the earlier version? What is the main improvement with the current one? Could these issues have been presented here at the talk page before they were implemented?
- Almost the complete text in section "State of the art" is removed. What was the reason for this?
- In section "Related Fields" there are now two paragraphs, 3rd and 4th, that discuss the relation between CV and biological vision. To me there seems to be a bit of overlap between the two.
- In section "Typical tasks of computer vision" there is both a "scene reconstruction" and a "3D volume recognition" task. What is the main difference between the two?
- Given that the article on CV presents an overview of the area, it seems to me that the "3D volume recognition" goes a bit too deep into the area, presenting concepts and statements without explaining. Why is this necessary?
--KYN (talk) 16:50, 8 October 2011 (UTC)
- Because different editors edit different parts of the article without necessarily reading the rest of the article and noting what is already included elsewhere.
- Welcome to Wikipedia .. blah blah .. -- Short version: If you don't like the way the article is written, click the Edit tab at the top and go to it.
- If 3D volume recognition section is too complex for your liking in a general article, then either move it to another new article or make sure the material already is documented in an existing article, before deleting it.
Hello KYN. I did many of the recent edits. It would take hours to try to try to answer your questions the way that you wrote them, and some have already been answered in this talk page a mere few posts back. But a few salient notes:
- Although I did lot of work, I'm certainly not going to go to bat for the article in it's current state. It needs a LOT of work, especially organization and removal of duplications.
- I didn't put the "real time" in there but I suspect it was put in because most work referred to as computer vision involves automated and continuous analysis of images, usually where the results are available and even used within a fraction of a second of the imaging. Off-line type analysis is usually not called computer vision, usually it's called image processing, or image analysis.
- One problem with the previous version which I have just barely started to fix is that it had ZERO references/citations.
- Per talk, I have my own concern about the 3D volume recognition section which is that it is not really explanatory. I asked the original inserter (both here and on their talk page) if they'd work on it further....so far no luck, but there's no hurry.
- IMHO (apologies if I'm offending anyone here) the previous "state of the art" section was a bunch of confusing rambling that said absolutely nothing.
Sincerely, North8000 (talk) 21:38, 9 October 2011 (UTC)
- Can you expand your comment to issue 6 above. I'd like to hear you opinion on that before suggesting what to do. --KYN (talk) 08:36, 13 October 2011 (UTC)
- As worded it would take a lot of tangential work (discussing other hypothetical ways it could have been done, explaining Wikipedia BRD norms, reinforcing a false premise that the only valid reason for improving an article is proving "wrongness" of the previous version, and doing an analysis for "wrongness" for the previous version. But in comparing the two over time, it looks like two (unsourced) errors were corrected (calling machine vision the simple type of CV, and saying that a typical MV application is counting bottles.) Another significant change was clarifying CV vs. MV. But it still needs more work. Others should keep improving it. Sincerely, North8000 (talk) 12:14, 21 October 2011 (UTC)
- Ok, I see. The type of improvement I had in mind was to revert the lead part of the article to the version from 2011-05-31T15:03:39, before any of the recent edits on it were made. That would take care of issues 1-6. Is there any important information in the current version that would be removed by such a revert? --KYN (talk) 13:58, 6 November 2011 (UTC)
- Several issues there. Wiping out 6 months of development on the lead isn't really the way to edit the article. North8000 (talk) 02:02, 7 November 2011 (UTC)
- My point is that there are several problems with the lead, listed above, which in my view boil down to: these last months of edits do not really represnt a development in a positive sense. Can you be more specific about what would be lost in a revert? --KYN (talk) 08:57, 7 November 2011 (UTC)
- That places a huge burden of work (analyzing two versions of the lead 6 months apart for what would be lost) to avoid the extraordinary and unusual action of undoing the gradual evolution of the lead over the last 6 months. I believe that the evolved version fixed problems and added important missing information. Sincerely, North8000 (talk) 11:43, 7 November 2011 (UTC)
- I have already done that analysis and came to the conclusion that a revert amounts to an improvement of the current state. If you disagree, I hope that you can point out one or a few things that really needs to be preserved or shouldn't be touched. Alternatively, since you are using a terms like "development" and "evolution" in a positive sense, what was missing before the last months of edits were made? --KYN (talk) 15:31, 7 November 2011 (UTC)
- It's not reasonable or normal to say that someone must spend hours to answer your question as you formatted it in order to stop a 6 month reversion of the entire lead. People have made 6 months of contributions to it subject to the normal practices, including tacit or explicit acceptance of the changes, and prior discussion of the ones where there might be a question, dea;lign with each item individually. (From a brief look, I can see that the newer one provides a scope which better matches the field and is more explanatory. Also the partially-overlapping field of machine vision is covered and put in context in relation to CV.) Future changes should follow this same process. Sincerely, North8000 (talk) 20:52, 7 November 2011 (UTC)
Reference method
The recent edits have changed the referencing method to a very unusual structure which may be confusing/difficult for many editors to use. Sincerely, North8000 (talk) 12:58, 30 November 2011 (UTC)
Diagram
The diagram has been tagged as OR for a long time. Also it basically says little or nothing......actually is a unsourced & wrong statement of non-relationships. It basically says that 18 listed disciplines have no relation to each other, and that 3 have a relationship. I plan to delete it from the article. North8000 (talk) 02:38, 23 February 2013 (UTC)
- Sorry, I missed the OR-tag completely. Where do we find that tag? If you want to remove things like that, better discuss it here first. About the diagram itself, I see it as an illustration of the text that accompanied it. This means: if the diagram is wrong, then the text is wrong. Can you please expland on this issue before removing the diagram? --KYN (talk) 16:50, 25 February 2013 (UTC)
- I removed it a few days ago but there's no rush. I'll put it back in so we can discuss. My main thoughts are in my Feb 23 post. Sincerely, North8000 (talk) 16:53, 25 February 2013 (UTC)
- Ok, I see. Let's start with the text next to the figure, which the figure is supposed to illustrate. Is there a problem there? --KYN (talk) 18:31, 25 February 2013 (UTC)
- No problem that I know of. I wasn't aware that that figure was intended for that particular text, but I consider that to be unusual, not to be a problem. North8000 (talk) 18:37, 25 February 2013 (UTC)
- I agree that the figure may not be straight-forward to decode without reading the text, although there is a caption that tries to make some connections, and from that point of view it may somehow be improved. Any suggestions? --KYN (talk) 21:25, 25 February 2013 (UTC)
- I don't know what to say because I don't know what it is trying to be. Normally something that looks like that is is a Venn diagram which shows any overlaps, subsets etc.. But reading this one that way, it shows 18 of the fields completely separated from each other which means that they have none of the relationships that the diagram is supposed to show. The three items in the middle are good. Sincerely, North8000 (talk) 03:31, 26 February 2013 (UTC)