Computer Vision and Machine Vision

Wikipedia, a free encyclopedia  

Computer VisionIs a study on how to make machinesViewThe science, to put it further, refers to the useCameraAndComputerTarget Recognition, tracking, and measurement instead of human eyesMachine VisionAnd furtherGraphic ProcessingAnd use a computer to process images that are more suitable for human observation or transmission to the Instrument for detection.

As a scientific discipline, theories and technologies related to computer vision research attempt to establish an artificial intelligence system that can obtain 'info' from images or multidimensional data. The information here refersShannonDefined to help make a "decision. Because perception can be seen as extracting information from sensory signals, computer vision can also be seen as a science that studies how to make human systems "perceive" from images or multidimensional data.

As an engineering discipline, computer vision seeks to establish it based on relevant theories and models.Computer Vision System. Components of such systems include:

1. Program Control (for exampleIndustrial RobotsAndSelf-driving car)
2. Event monitoring (for exampleImage Monitoring)
3. Information Organization (indexing of image databases and image sequences)
4. Object and Environment Modeling (such as industrial inspection, medical image analysis, and Topology modeling)
5. Sympathetic interaction (such as human-computer interaction input device)

Computer Vision can also be seen as a supplement to Biological vision. In the field of Biological vision, human vision and various animal vision have been studied, so as to establish the physical models used by these visual systems to perceive information. On the other hand, artificial intelligence systems based on software and hardware have been studied and described in computer vision. Communication between biological vision and computer vision brings great value to each other.

Computer Vision includes the following branches: Image Reconstruction, event monitoring, target tracking, target recognition, machine learning, indexing, and image restoration.

1 Development Status of Computer Vision 2 similarities and differences between adjacent fields 3 typical computer vision problems 3.1 Recognition 3.2 exercise 3.3 scenario Reconstruction 3.4 Image Restoration 4. Computer Vision System 5 requirements affecting the visual system 6. Reference 7. External links

Development Status of Computer Vision

Relationship between computer vision and other fields

Computer Vision is characterized by its diversity and imperfection.

The pioneers in this field can be traced back earlier, but20th centuryIn the late 1970s s, when the computer's performance was improved to be sufficient to handle suchImageComputer Vision has been officially followed and developed for such large-scale data. However, these developments often originate from the needs of other different fields. Therefore, the definition of "computer vision problems" has never been formally defined. Naturally, there is no formula for solving the "computer vision problem.

Even so, people have begun to master some methods to solve specific computer vision tasks. Unfortunately, these methods are generally only applicable to a group of narrow targets (such as faces, fingerprints, and texts ), therefore, it cannot be widely used in different scenarios.

The application of these methods is usually used as a solution.Complex problemsAn integral part of a large-scale system (such as medical image processing, quality control and measurement in industrial manufacturing ). In most practical applications of computer vision, computers are preset to solve specific tasks. However, machine learning-based methods are becoming increasingly popular, in the future, "generic" Computer Vision Applications may come true.

Artificial IntelligenceOne of the main problems studied is: how to make the system "plan" and "decision-making ability "? So that it can complete specific technical actions (for example, moving a robot through a specific environment ). This issue is closely related to computer vision. Here, the computer vision system asSensorTo provide information for decision-making. Other research areas includePattern RecognitionAndMachine Learning(This is also part of the AI field, but it has an important relationship with computer vision.) As a result, computer vision is often seen as artificial intelligence andComputer Science.

PhysicalIt is another area that has an important relationship with computer vision.

The goal of computer vision is to fully understandElectromagnetic Wave-- MainlyVisible LightAndInfraredPart -- encounters an object surfaceReflectionThe image, and this process is based onOptical PhysicsAndSolid state physics, Some cutting-edgeImage sensing systemIt will even be appliedQuantum mechanicsTheory to parse the real world represented by images. At the same time, many measurement problems in physics can also be solved through computer vision, suchFluidExercise. Therefore, computer vision can also be seen as an extension of physics.

Another important field isNeurobiologyIn particularBiological Vision System.

Throughout the 20th century, humans have extensively studied the eyes, neurons, and brain tissues related to visual stimulation of various animals, these studies have produced some descriptions about how a "natural" visual system works (though not too rough ), this also forms a sub-field in computer vision-people try to build human systems to simulate the visual operation of biology to varying degrees of complexity. In the field of computer vision, some machine learning-based methods also refer to some biological mechanisms.

Another field related to computer vision isSignal Processing. Many problemsUnit variable Signal, EspeciallyTime-varying signalCan be naturally extended toBinary variable SignalOrMultivariate Signal. HoweverImage DataMany methods developed in computer vision cannot find the corresponding version in the unit signal processing method. One of the main features of these methods is theirNon-linearAnd Image InformationMultidimensionalAs part of computer vision, the above two points form a special research direction in signal processing.

In addition to the fields mentioned above, many research topics can also be regarded as pureMathematicsProblem. For example, the theoretical basis of many problems in computer vision isStatistics,OptimizationTheory andGeometry.

How to implement existing methods through various software and hardware, or how to modify these methods to achieve a reasonable execution speed without losing enough precision is the main topic of today's computer vision field.

Similarities and differences between adjacent fields

Computer Vision,Image Processing,Image Analysis,Robot VisionAndMachine VisionIs a closely related discipline. If you open a textbook with the above names, you will find that they overlap a considerable part of the technology and application fields. This shows that the basic theories of these disciplines are roughly the same, and even people suspect that they are named differently for the same discipline.

However, research institutions, academic journals, conferences, and companies often classify themselves into a specific field, so various characteristics used to distinguish these disciplines are proposed. A differentiation method is provided below, although it is not possible to say that this method is completely accurate.

Computer VisionThe research object is mainly mapped to a single or multiple images of 3D scenes, such as reconstruction of 3D scenes. The Research on Computer Vision is very specific to the image content.

Image ProcessingAndImage AnalysisThe research object is mainly two-dimensional images to achieve image conversion, especially for Pixel-level operations, such as improving image contrast, edge extraction, de-noise and geometric conversion image rotation. This feature shows that the research content of image processing and image analysis is irrelevant to the specific content of the image.

Machine VisionIt mainly refers to the Visual Research in the industrial field, such as the Vision of autonomous robots, used to detect and measure the vision. This indicates that in this field, image sensing and control theories are often closely integrated with image processing through software and hardware to achieve efficient robot control or various real-time operations.

Pattern RecognitionVarious methods are used to extract information from signals, mainly using statistical theories. A major direction in this field is to extract information from image data.

Another field is calledImaging Technology. The initial research in this field mainly involves image production, but sometimes involves image analysis and processing. For example,Medical ImagingIt involves a large number of image analysis in the medical field.

One possible process in all these fields is that you work in a computer vision lab, work in image processing, and finally solve the problems in the Machine Vision Field, then, we posted our results at the pattern recognition conference.

Typical Computer vision problems

Almost every computer vision technology has to solve a series of identical problems. Typical problems include:

Recognition

A typical problem of computer vision, image processing, and machine vision is to determine whether a group of image data contains a specific object, image features, or motion status. This problem can usually be solved automatically by machines, but so far, there is no single method that can be widely used to determine various situations: To identify arbitrary objects in any environment. The existing technology can only well solve the recognition of specific targets, such as simple geometric image recognition, face recognition, printed or handwritten file recognition or vehicle recognition. In addition, the recognition requires a specified illumination, background, and target posture in a specific environment.

Generalized recognition has evolved into several slightly different concepts in different scenarios:

• Identification (narrow sense ):One or more objects or objects that have been pre-defined or learned are identified. Generally, their two-dimensional positions or three-dimensional attitudes are also provided during the identification process.

• Identification:Identifies and identifies a single object. For example, recognition of a face and recognition of a fingerprint.
• Monitoring:Discover specific content from the image. For example, the discovery of abnormal cell or tissue skills in medicine and the discovery of vehicles by traffic monitoring instruments. Monitoring often discovers Special Areas in images through simple image processing, providing a starting point for more complex operations.

Specific application directions:

• Content-based image Extraction:Search for all images containing the specified content in a huge image set. The specified content can be in multiple forms, such as a red circle or a bicycle. Here, the search for the latter content is obviously more complex than the previous one, because the former describes a low-level visual feature, the latter involves an abstract concept (or an advanced visual feature), that is, a 'bicycle '. Obviously, the bicycle's appearance is not fixed.

• Pose evaluation:Evaluate the position or direction of an object relative to the camera. For example, evaluate the posture and position of the machine arm.

• Optical Character RecognitionRecognition and identification of printed or handwritten text in an image. The output is usually converted into a document that is easy to edit.

Sports

The monitoring of object motion based on sequential images involves multiple types, such:

• Auto motion:Monitors three-dimensional rigid motion of the camera.
• Image tracking:The object that tracks motion.

Scenario Reconstruction

Given two or more images or a video clip of a scenario, scenario reconstruction seeks to createComputer Model/3D Model. The simplest case is to generate points in a group of 3D spaces. In more complex cases, a complete 3D surface model is created.

Image Restoration

The goal of image restoration is to remove noise from the image, such as instrument noise and blur.

Computer Vision System

The structure of a computer vision system depends largely on its specific application direction. Some work independently to solve specific measurement or detection problems, and some appear as components of a large complex system, such as mechanical control systems, database systems, human/machine interface devices work collaboratively. The specific implementation methods of computer vision systems are also determined by their functions-whether they are pre-fixed or automatically learned and adjusted during running. However, some functions are required by almost every computer system:

• Image Acquisition:A digital image is generated by one or more image sensors. The sensor can be a variety of photosensitive cameras, including remote sensing devices, X-ray tomography, radars, and ultrasonic receivers. Instead of the same sensor, the images produced can be ordinary two-dimensional images, three-dimensional graphs, or an image sequence. The pixel values of an image often correspond to the intensity of light on one or more spectral segments (grayscale or color graphs), but they can also be related to various physical data, such as acoustic waves, electromagnetic waves, orMRIDepth, absorption or reflection.

• Preprocessing:Before a specific computer vision method is used to extract specific information from an image, one or more preprocessing methods are often used to meet the requirements of subsequent methods. For example:
  • Secondary Sampling ensures correct image coordinates
  • Smooth noise removal to filter out device noise introduced by Sensor
  • Increase the contrast to ensure that the information can be detected.
  • AdjustmentScale SpaceMake image structure suitable for local applications
• Feature Extraction:Extract features of various complexity from the image. For example:
  • Line,Edge Extraction
  • LocalizedFeature DetectionFor exampleCorner Detection,Spot Detection

More complex features may be related to the texture shape or motion in the image.

• Detection/segmentation:In the process of image processing, it is sometimes necessary to divide the image to extract valuable parts for subsequent processing, such
  • Filter Feature Points
  • Split one or more images with specific targets
• Advanced processing:In this step, the data is often very small. For example, we can pre-process the part that is considered to contain the target object first. At this time, the processing includes:
  • Verify that the data meets the prerequisites
  • Estimate specific coefficients, such as the target attitude and volume
  • Classify targets

Elements that affect the visual system

• Careful consideration is required for the impact of light source layout.
• Select the mirror Group correctly, consider the magnification, space, size, distortion... .
• Select a suitable camera (CCD) and consider functions, specifications, stability, and durability ....
• Visual Software Development requires accumulation of experience, multiple attempts, and solutions to problems

Channels.

• The ultimate goal is to continuously improve the precision and shorten the processing time.

Reference

• Machine Vision
• Artificial IntelligenceAndPattern Recognition
• Image Processing
• Automatic Optical Inspection
• Open-source computer vision Library: opencv

External link

• Machine Perception of Three-dimen1_solids-the paper mentioned by Joseph Mundy in the video
• Cvonline: the evolving, distributed, non-proprietary, on-line compendium of Computer Vision
• Introduction to Computer Vision(464kb PDF File)
• CMU'sComputer Vision Homepage
• Keith Price's annotated Computer Vision BibliographyAnd the official mirror siteKeith Price's annotated Computer Vision Bibliography
• Hipr2 Image Processing Teaching Package
• USC Iris Computer Vision Conference List

From"Http://zh.wikipedia.org/w/index.php? Title = % E8 % AE % a1 % E7 % AE % 97% E6 % 9C % Ba % E8 % A7 % 86% E8 % A7 % 89 & variant = ZH-CN"