Talk:Computer vision/Archive 1

Latest comment: 16 years ago by 24.61.184.39 in topic History
Archive 1

biological vision


Interestingly, biological vision (cats have been extensively studied) seems to have lots of "hardware" assist, that is, special "circuitry" to find lines, etc. However, most computer vision has been done strictly with software.

Removed. This implies that there is a clearly understood, meaningful distinction in neuroscience between hardware and software in the brain. Since this is not the case, I don't see how we can include this.

Also, note that things like the "special 'circuitry' to find lines" have been copied into computer vision efforts; they're called "feature extractors", or some such. --Ryguasu 21:51 Dec 7, 2002 (UTC)


Much machine vision--which is applied computer vision--takes advantage of specialized hardware to pre-process results. In the 1980s and 1990s it was not uncommon for machine vision equipment manufacturers to develop specialized hardware to perform common operations (esp. connectivity analysis) on incoming video signals. More recently, specialized hardware has been replaced by configurable, programmable boards capable of processing input before the image is presented to a higher level of interface accessible to the programmer or user. -- Rethunk

This article is almost as Machine vision..

Should Machine vision and Computer vision combined?


I removed "Computer vision is a subfield of artificial intelligence and computer graphics." Computer graphics and computer vision have very separate histories, and only within the last 10 years have people started to explore the overlap within the fields. Computer vision arose from artificial intelligence, and was original a sub-topic of one of the main AI conferences (IJCAI, I think). Roughly speaking, computer vision and computer graphics address related, but distinct problems. Computer graphics is about converting scene descriptions to images, and computer vision is about converting images to scene descriptions; one is not a subproblem of the other. In both research and practice, the fields have historically been separate, but this is changing.

I would argue that "machine vision" is a synonym for "computer vision." However the content in the machine vision article is a very limited view of computer vision that focuses on primitive techniques that would not be considered broadly representative of computer vision by any means. -- Aaronh

The content of the "machine vision" article focuses on some simple processing techniques, but for those unfamiliar with the field this information could be useful. The article also (briefly) mentions some algorithms that anyone with a broad grasp of computer/machine vision recognizes as hardly "primitive". The article does not appear to claim that "machine vision" encompasses the same scope as "computer vision"; in fact, in the article's current iteration it describes machine vision as a subfield of computer vision. This assertion dates back to the first version of the article. Thus the comments about the "primitive techniques" that are "not . . . broadly representative of computer vision . . . by any means" seem besides the point (and also inappropriate in tone from someone who is a professional in the field). -24.61.184.39 (talk) 07:07, 19 November 2007 (UTC)

Tracking

I removed the "Possible terrorist tracking in the airport" in the examples of tracking tasks. I have never heard of a system which can do this. If someone has a reference to a working system which has solved the terrorist detection, we can put it back in again. KYN 23:41, 12 November 2005 (UTC)

A wrong pointer / copy of this page?

The people from a company called Riya claim that the computer vision page on wikipedia is completely different from what it actually is... Please check http://www.riya.com/corp/history-face-recongnition.jsp . I think this is not very fair, as this false content is a pure publicity for Riya, which by the way is a very young new comer in the field of face recognition, that does not even test on the FERET db or do the FRVT test.

Should we contact them to correct that? --- They are likely referring to the italicized quote: "Computer vision can be described as the study of methods which can be used for allowing computers to "understand" images, or multidimensional data in general." -varaon

Suggestions for improvements of the Computer Vision article and category

Please read the suggested improvement posted at The Category:Computer Vision Talk page. --KYN 11:38, 5 February 2006 (UTC)

Introduction

The introduction to this article seems to be its current weak point. I'm not sure I want to mess with it yet for fear of leaving something out or including my own possible misconceptions or ambiguities, but right now it leaves the reader a bit confused. The first paragraph is fine. The second seems a bit irrelevant at this point in the article; perhaps it should be moved to another section? The third part (point form) should probably be removed; weak grammar aside, it seems useless and redundant. Doze 20:14, 09 May 2006 (UTC)

I agree that the introduction of the article lost clarity when the third paragraph appeard. The second paragraph is not irrelevant but possibly misplaced. Here's my suggestions:
  • Delete paragraph 3
  • Delete paragraph 2, and write its content into paragraph 4 of state-of-the-art section
--KYN 19:59, 16 May 2006 (UTC)

Computer vision systems

This section seems a bit unstructured or unorganized. I suggest to replace the content of this section with something more concise like:

The organization of a computer vision system is highly application dependent. Some systems are stand-alone applications which solve a specific measurement or detection problem, while other constitute a sub-system of a larger design which, for example, also contains sub-systems for control of mechanical actuators, planning, information databases, man-machine interfaces, etc. The specific implementation of a computer vision system also depends on if its functionality is pre-specified or if some part of it can be learned or modified during operation. There are, however, typical functions which are found in many computer vision systems.

  • Image acquisition: A digital image is produced by one or several image sensor which, besides various types of light-sensitive cameras, includes range sensors, tomography devices, radar, ultra-sonic cameras.
  • Pre-processing: Before a computer vision method can be applied to image data in order to extract some specific piece of information, it is usually necessary to process the data in order to assure that it satisfy certain assumptions implied by the method. Examples are
    • Re-sampling in order to assure that the image coordinate system is correct.
    • Noise reduction in order to assure that sensor noise does not introduce false information.
  • Feature extraction: Image features at various levels of complexity are extracted from the image data. Typical examples of such features are
    • Lines and edges.
    • Localized interest points such as corners or points.
More complex features may be related to texture, shape or motion.
  • Segmentation: At some point in the processing a decision is made about which image points or regions of the image are relevant for further processing. Examples are
    • Selection of a specific set of interest points
    • Segmentation of one or multiple image regions which contain a specific object of interest.
  • High-level processing: At this step the input is typically a small set of data, for example a set of points or an image region which is assumed to contain a specific object. The remaining processing deals with, for example,
    • Verification that the data satisfy model-based and application specific assumptions.
    • Estimation of application specific parameters, such as object pose or object size.
    • Classifying a detected object into different categories.

I will change to this content shortly, if there are no major objections. The current content is of course still relevant to computer vision, but is too specific in my opinion. Maybe we can have a separate article or category which in a meaningful way can list various types of processing methods which are frequently appearing in computer vision. --KYN 10:19, 2 August 2006 (UTC)

New introduction

I would like to suggest the following alternative:

Computer vision is the science of machines that see.

As a scientific discipline, Computer vision is concerned with the theory and technology for building artificial systems that obtain information from images or multi-dimensional data. Information, as defined by Shannon, is that which enables a decision. Since perception can be seen as the extraction of information from sensory signals, computer vision can be seen as the scientific investigation of artificial systems for perception from images or multi-dimensional data.

Examples of applications of ((computer vision systems)) include systems for

  1. Controling processes (e. g., an industry robot or an autonomous vehicle).
  2. Detecting events (e.g., for visual surveillance for security or commercial services)
  3. Organising information (e.g., for indexing data bases of images and sequences),
  4. Modeling objects or environments (e.g., industrial inspection, medical image analysis, or topographical modeling),
  5. Interaction (e.g., as the input to a device for computer-human interaction).

Computer vision can also be described as the complement (but not necessary the opposite) of biological vision. In biological vision and visual perception real vision systems of humans and various animals are studied, resulting in models of how these systems are implemented in terms of neural processing at various levels. Computer vision, on the other hand, studies and describes artificial vision system which are implemented in software or hardware, in computers or in embedded systems. Interdisciplinary exchange between biological and computer vision has proven increasingly fruitful for both fields.

Subdomains of computer vision include scene reconstruction, event detection, tracking, object recognition, learning, indexing, ego-motion and image restoration.

-JLC

This looks fine to me. What maybe is missing from the current introduction is that computer vision can deal with many types of image sensors, not just standard 2D sensors, and that learning-based systems are becoming useful. The last issue can maybe be worked into the first paragraph of the "State of the art" section and the first one worked into the "Computer vision systems" sections. Also, my spell-checker says "Controling->Controlling", but it has been wrong before... --KYN 14:20, 3 August 2006 (UTC)
Well, one more thing. If you write that computer vision is a science then certain people are inclined to interpret that it is only about research, not about real world applications. --KYN 19:21, 3 August 2006 (UTC)
ok. I agree. I have cleaned up the text, and used American, rather than British spelling.

Scene reconstruction

I removed the last sentence

"Generally one is interested in finding the fundamental matrix which represents corresponding points from different images. "

from this section. I) The fundamental matrix is a means, not the goal of scene reconstruction. II) I don't believe that this article is a right place to describe the specific methods for solving the scene reconstruction problem. --KYN 17:02, 30 October 2006 (UTC)

Remove Computer vision laboratories list

There is a list of CV labs at the end of the article. It is a result of a reorganization of the article made about a year ago, where things were moved around rather than deleted, which perhaps they should have. I have noticed that lately a number of new labs have been added to this list and, given time, it could soon be as long as the rest of the article. Furthermore, this list is not, and can probaly never be, a complete list of CV labs. I have tried to find similar "lab lists" in other articles, but so far didn't find any so I guess that this list is rather unique for the CV article. For these reasons I propose to remove the list from the article. --KYN 11:57, 14 November 2006 (UTC)

A better solution might be to create a Computer_vision_labs page with a complete list and more information about each lab like their specific focus and past contributions to the field. --Dougthebug 19:18, 19 January 2007 (UTC)

OK, let's move the list to a "Computer Vision Research Groups" page for a start. However, I reject the idea that the list will ever be "complete", which is one of the reason why I want to remove it from the CV main page. Also, I would be careful to use Wikipedia to duplicate information that is already presented on each group's home page. --KYN 12:45, 30 January 2007 (UTC)

Done! --KYN 20:05, 15 February 2007 (UTC)

I think this link http://www.cs.ubc.ca/spider/lowe/vision.html should be added to the external resources list, as it includes a list to many industrial applications of computer vision. --Hl

Move "computer vision systems" info into separate article

I propose that the following be moved to the computer vision systems page.

  1. paragraph two - "Examples of applications...thru item 5"
  2. section titled "Vision based biological species identification systems"
  3. section titled "computer vision systems"

The purpose is to allow full development of the two concepts (computer vision and computer vision systems) in independent articles. If there is no disagreement I will make this change later this week. --Axiomatica 18:55, 19 June 2007 (UTC)

Moving the compute vision systems section to a separate article may be a good idea, in particular in you plan to expand it. However, I disagree on the first two points above.
The section Examples of application for computer vision is important for the CV article, without it the reader has no idea what CV can be used for. Note that this section is only indented for providing a short overview of examples of what CV can be used for, it is not an exhaustive presentation. If you plan to do something like that, that could be an article by itself, but I still want to keep the mentioned section in the CV article as an overview (it may be shortened if there was a longer application article).
About the second point, I get the impression that you want to mix/combine computer vision applications and computer vision systems. I'm not saying that this cannot be done, but there is a problem of taxonomy here. To me applications and systems have different meanings, although I not certain that I can present a clear and concise definition with also everybody else agrees with.
To give you an idea of my own (not universally accepted) taxonomy looks like
  • A computer vision application is a high level problem which happens to be solvable by applying computer vision techniques, but in general also requires techniques from other areas, such as automatic control, (general) computer science (e.g., autonomous reasoning) or information processing (e.g., GIS). Examples of this type of applications are autonomous vehicles which requires a larges set of techniques other than computer vision to solve the problem.
  • An application is typically broken down into tasks (sub-problems) which need to be solved, and now we focus on the tasks which relate to computer vision. For example, in the case of autonomous vehicles, typical tasks are detection of obstacles, where they are and how they are moving, building maps of the environment, determining the vehicle's position relative to some map, etc.
  • Given a specific task, it can normally be solved by means of various methods or techniques (also popularly referred to as algorithms). There are numerous methods for detecting objects, for motion estimation, for stereo triangulation, etc.
  • Given a specific method, it can be implemented in various ways, in various hardware and software configurations, in dedicated hardware, on a graphics card, etc. In my view, the implementation aspect of computer vision comes very close to what I would refer to as a computer vision system.
In many cases, a computer vision system is application specific, which makes the connection between applications and systems reasonable. However, since some computer vision systems are general and can be programmed to solve one of several applications, I believe that they shouldn't be mixed. In short, it makes sense to describe and discuss general aspects of computer vision system without referring to a specific application, and vice versa. --KYN 22:36, 5 July 2007 (UTC)
By the way, I agree that the section titled "Vision based biological species identification systems" should be moved to somewhere else. It is rather excessive and technical to fit the rest the of the article. My objection was simply that "Applications" rather than "Systems" would be a better heading for this material. --KYN 22:39, 5 July 2007 (UTC)

To deal with the section entitled "Vision based biological species identification systems" I propose to move it to a new article with the same title and link it to the CV category. However, since there are no proper references to anything inside or outside WP it could also be removed altogether. Any opinions? --KYN 20:56, 25 July 2007 (UTC)

I'v removed the section "Vision based biological species identification systems" from the CV article. Ifs someone wants it back, please put it in a separate article with proper internal and external links/references. --KYN 08:23, 1 August 2007 (UTC)

You have made excellent improvements to this article, KYN. I am sorry to see the "Vision based biological species identification systems" disappear entirely, I'll put it on my list to revisit and see if there's enough there to expand into a full entry.--Axiomatica 18:49, 18 September 2007 (UTC)

I removed the "Vision based biological species identification systems" section with mixed feelings. On one hand its it a reasonable example of an application computer vision application, but on the other hand it was too long and too detailed to stay. However, making a full article of it with proper refs, links, etc sounds like a great idea. I guess that in the end, there could be a "CV applications" article which develops this theme to a greater extent than is possible in the main article. --KYN 21:28, 18 September 2007 (UTC)

The "See also" section

This section is a sort of garbage collection of various links which was found in an earlier version of the article. Both to subcategories of the computer vision category and to "related articles".

Problem #1: Having a list of subcategories here is probably not a good idea since it needs to be updated each time someone adds a subcategory to the CV category (unless there is some automagic way to make that happen. Having the same information at several places is usually not a good idea and its seems to be a better idea to market a link to the CV category which also presents its subcategories.

Proposed solution: remove the subcategory list.

Problem #2: Having a list of "related articles" also appears to be difficult. The original idea was to maintain a list of wikilinks to articles which are "at the same level" as the CV article. This means technology or application areas which are in some sense at the same level as computer vision is, e.g., image processing, machine vision, or artificial intelligence, etc. Over time, however, the list has been extended with various articles which indeed are related to CV but sometimes are referring to specific methods or applications rather than areas of such things. In the end, this list may grow to something similar to what is already found in the CV category.

Proposed solution: remove wikilinks to articles which either is not a larger field of technology, research, or applications, or if it is can be said to be part of another field which is already in the list.

For the current list this could give the following result

--KYN 21:43, 25 July 2007 (UTC)

Done! --KYN 09:05, 1 August 2007 (UTC)

I commented out the second tutorial as the link is broken. There are more tutorials and white papers out there that would be appropriate for this section if anyone has an interest in tracking them down.--Axiomatica 19:20, 18 September 2007 (UTC)

History

I think a discussion of the history of the field would be a good improvement to the article. When did this field originate? What were the first "computer vision" applications? What were the earliest problems tackled under the name of computer vision? These could be starting points to a history section. Sancho 07:04, 18 October 2007 (UTC)

To write a proper historic presentation of this field is not an easy task. For one thing there appears to be very little written about this. I have about 10 basic computer vision text books and most of them avoid the historic presentation entirely, in fact only two of them say anything at all. Tim Morris' book has a short section (1.2) and Gonzalez & Woods (3rd ed) has a rather extensive historic presentation (sec 1.2). But none of these books can trace what we call "image processing" or "computer vision" to individual persons or research groups, it is more in the context of general technical developments. What I'm saying is that it is difficult to write a verifiable historic presentation. --KYN 11:36, 18 October 2007 (UTC)
I know there are some review papers included as an introduction in some paper collections. I've seen some historical summaries in these. Also, here's what I've found so far:
  • Lecture slides that include a history section
  • Zenon W. Pylyshyn. Seeing and Visualizing: It's Not What You Think. ISBN 0262162172 (Pages 58-61 give a brief overview of the history of the field as it parallels advances in understanding of the biological aspects of vision)
  • Another broad overview (one sentence about each decade since the 60s)
  • Azriel Rosenfeld. Image Analysis: Problems, Progress and Prospects. In Readings in Computer Vision: Issues, Problems, Principles, and Paradigms, 3-12, 1987.
Even if we're only able verifiably talk about trends and general focus of research throughout the history of this field, that would be good I think. Sancho 02:51, 19 October 2007 (UTC)
If you want to write something like a historic exposé, just go ahead and do it. You just have to be prepared that there is likely to be a bunch of people out there who are going to worry about the exact formulations on who did what and when. On the other hand, isn't that the very idea of Wikipedia? --KYN 16:43, 3 November 2007 (UTC)
Full agreement with KYN's comment: go for it, and good luck! A few algorithmic techniques could be traced relatively easily to individuals (e.g. Hough, Duda, Hart) or to groups (SRI). I recall that a chapter or two in a book by Nello Zuech covered the history of some vision techniques used for automated inspection in manufacturing. You may find that the only thing resembling a proper "history" of the field would exist in the brains of those who practice it. It appears that Hart of "Duda and Hart" fame posted in the Talk section of the Hough transform article--maybe some of the best sources for info are occasional Wikipedia contributors. -24.61.184.39 (talk) 07:36, 19 November 2007 (UTC)