Computer Vision

Computer vision fulfils a dual objective, biological and engineering. As far as the biological point of view is considered, computer vision approach enables different models of human visual system. While from the perspective of engineering, the approach enables developing an autonomous system which has the potential to perform few tasks that can be performed by human visual system. A majority of the vision tasks are associated using time independent 2D data for obtaining 3D and temporal information. The 2D data is retrieved from television, cameras etc.

Larry Roberts, known as the father of Computer Vision described the probability of extracting 3D geometrical data from 2D perspective view (Shapiro, 1992) . However, it was later identified by the researchers that it is imperative to employ images from the real world. Hence, there was still needed some more research to be performed in the tasks such as edge detection and segmentation. These tasks are categorized as low-vision tasks. In the field of computer vision, one of the most significant works was performed by David Marr who proposed following bottom-up approach to scene understanding (Marr, 1982).

Taking into consideration the historical achievements made in the field of computer vision, it can be asserted that due to the large-scale employability of the approach computer vision be integrated with some other related fields such as:
Image processing: It refers to processing of raw images for performing further analysis.
Photogrammetry: It emphasizes on the necessity of calibration of cameras that are to be employed for imaging.
The field of computer vision is a highly complex field and thus there has not been resolved any research issue yet. The main reason for complexity in resolving the issues is that human visual system exhibits better performance for several tasks such as face recognition and thus computer vision has to suffer the consequences of comparison between the systems. For example, a human visual system can identify faces even under different kind of variations in illumination, expressions etc.

In majority of the cases there is no complexity in recognizing the face of an individual in a picture that was taken several years ago. Additionally, the number of faces that can be stored by our brain for future recognition is also not confined. Such performance cannot be exhibited by the autonomous or computer vision systems developed. Thus the two major complexities encountered in computer vision systems are as follows:
• How to perform filtering and representation of the vast human knowledge in a computer in a manner that it can be retrieved easily?
• How to make large scale computation which is mostly needed in such a way that it can be done in real time?
The primary challenge associated with computer algorithms are that they are up to large extent brittle which implies that an algorithm might be brittle in some conditions but not in others. Thus, for an application based on computer vision to be successful there should be fulfilled two criteria:
• Possibility of human interaction
• Forgiving
Apart from fulfilling these criteria, it is also noteworthy that in some tasks the approach of vision feature must be combined with certain other characteristics such as audio. This integration with other features help in exhibiting better performance.
Thus considering the two criteria there are some applications of computer vision that can achieve success, for example:
Image/video databases- Image content-based indexing and retrieval.
Vision-based human computer interface- In this application gesture is integrated with speech for facilitating communication with virtual environments.
Virtual agent/actor- Through this application scenes of a person are produced on the basis of the parameters obtained derived from videos of real individuals.
Some of the key real world applications of computer vision are as follows:
Optical character recognition (OCR): This application of computer vision enables reading the postal codes on letters.
Machine inspection: This is the most significant application as it enables inspection of machine parts by employing stereo vision with special emphasis on estimating tolerance on aircraft wing parts.
Photogrammetry: Through this application 3D models are developed from the aerial pictures that are used in various systems, for example Bing Maps.
Medical imaging: The application of computer vision for medical or healthcare field is mostly preferred for conducting long term studies of brain morphology of people.
Automatic safety: This application aims at identifying obstacles that are not expected, for example pedestrians on the street.
.