TAPe-technology - a new word in image recognition
Greetings, dear Habrazhiteli!
Allow me to bring to your attention the Comexp company’s blog, as well as the TAPe technology, which allows you to recognize images very quickly, even in conditions of strong interference.
Interesting? I ask under the cat
')
As you know, there are some standard image processing methods that can cope with a fairly narrow class of technical problems. The tools used for image analysis can work only in conditions of minimal interference. If the interference becomes significant, then the reliability of image recognition is sharply reduced. This significantly increases the time of its processing, which negates all the tasks of detecting an object or tracking it. Still, the object disappeared a long time ago, and the system only after half an hour reported that it recognized it!
Classical image processing technology involves three main steps:
Prior to the development of TAPe technology, the first two stages did not have a theoretical solution, subject to the presence of organized interference. This is precisely the reason for the lack of algorithms for reliable recognition of a person from his photograph, or television images in the control systems of various law enforcement agencies.
Maybe this is a fiction, and it is impossible to create such a mechanism? However, here, as always, nature itself challenges us, once again demonstrating its perfection. A person’s visual analyzer is so perfect that we can recognize an object not only in difficult weather conditions, but also when most of this object is hidden from us. If it is not difficult for us to identify, for example, a glass by one edge, then for a machine it is an impossible task! While the machine works with a huge number of disconnected numbers, a person has a holistic perception, recognizing an object in large fragments by a small number of signs.
However, a solution to this problem was found. The company Comexp invented and patented such a technology that allows you to recognize an object, even if a significant part of it is blocked. This technology is called TAPe (eng. Theory of Active Perception - the theory of active perception).
The absence of convolution made it possible to recognize the object in question much more efficiently than was the case with pyramidal visual machines based on convolution. For example, pyramidal visual machines are 10 to 6 times more efficient than matrix machines!
You can look at the technology in action on our website in the “Test” section, there are various video clips that are viewed in real time from the recordings of weekly broadcasts of several TV channels in different regions. We change the list of TV channels and regions for testing every week; the current list is always displayed in the “Test” section. I want to note that in order to reduce the load on the database, we enabled caching on the site, so the report for each fragment is generated for the first time after its addition.
You can also upload your video (you will need to register in two steps) and see how quickly the database will be searched. If you do not have any videos from the air, we have prepared a list of current commercials and screensavers for testing.
Commercials:
Screensavers TV shows and TV shows:
I would like to know the opinion of our customers about our technology. Now we are developing the direction of monitoring television, as a priority in the use of TAPe technology, but in parallel we are developing an API for third-party developers. Are there any ideas on this subject, wishes, suggestions? I will be glad to know your opinion on this issue.
Thanks for attention.
Allow me to bring to your attention the Comexp company’s blog, as well as the TAPe technology, which allows you to recognize images very quickly, even in conditions of strong interference.
Interesting? I ask under the cat
')
Why do you need it?
As you know, there are some standard image processing methods that can cope with a fairly narrow class of technical problems. The tools used for image analysis can work only in conditions of minimal interference. If the interference becomes significant, then the reliability of image recognition is sharply reduced. This significantly increases the time of its processing, which negates all the tasks of detecting an object or tracking it. Still, the object disappeared a long time ago, and the system only after half an hour reported that it recognized it!
Classical image processing technology involves three main steps:
- Primary processing
- The allocation of classification features
- Image classification.
Prior to the development of TAPe technology, the first two stages did not have a theoretical solution, subject to the presence of organized interference. This is precisely the reason for the lack of algorithms for reliable recognition of a person from his photograph, or television images in the control systems of various law enforcement agencies.
Maybe this is a fiction, and it is impossible to create such a mechanism? However, here, as always, nature itself challenges us, once again demonstrating its perfection. A person’s visual analyzer is so perfect that we can recognize an object not only in difficult weather conditions, but also when most of this object is hidden from us. If it is not difficult for us to identify, for example, a glass by one edge, then for a machine it is an impossible task! While the machine works with a huge number of disconnected numbers, a person has a holistic perception, recognizing an object in large fragments by a small number of signs.
However, a solution to this problem was found. The company Comexp invented and patented such a technology that allows you to recognize an object, even if a significant part of it is blocked. This technology is called TAPe (eng. Theory of Active Perception - the theory of active perception).
The stages of the technology are as follows:
- Drawing up the initial description. One-time processing of the entire image. The absence of a convolution procedure at this stage avoids dependence on possible interference (noise). The arising interference becomes a part of the image, which can be used in favor of the object in question: in some cases it helps to get its informative elements; all existing pictures are identically displayed on a certain set of pictures, which can then be processed in real time using standard computers. For example, if we are talking about image processing on a field of 1024x1024 with brightness gradation levels of 256, then the power of a variety of images (the whole set of images) is a cosmic number of 256 1048576 . When processing, the power of the set of half-tone images is significantly reduced and can be represented by a small number of signs that carry information about the most significant structural elements of the image.
- Image analysis. Processing of the results of the first stage, spatial differentiation, which allows revealing the composite and structural connections between all the structural parts of the image.
- Image creation Special descriptive descriptions are prepared, reflecting the skeletons and integral curves of the structural connections of the image elements.
- Making decisions. Compiled graphic description is compared with the standard available in the database. If there is no standard, a new standard is formed.
The absence of convolution made it possible to recognize the object in question much more efficiently than was the case with pyramidal visual machines based on convolution. For example, pyramidal visual machines are 10 to 6 times more efficient than matrix machines!
View in action
You can look at the technology in action on our website in the “Test” section, there are various video clips that are viewed in real time from the recordings of weekly broadcasts of several TV channels in different regions. We change the list of TV channels and regions for testing every week; the current list is always displayed in the “Test” section. I want to note that in order to reduce the load on the database, we enabled caching on the site, so the report for each fragment is generated for the first time after its addition.
You can also upload your video (you will need to register in two steps) and see how quickly the database will be searched. If you do not have any videos from the air, we have prepared a list of current commercials and screensavers for testing.
Commercials:
Screensavers TV shows and TV shows:
I would like to know the opinion of our customers about our technology. Now we are developing the direction of monitoring television, as a priority in the use of TAPe technology, but in parallel we are developing an API for third-party developers. Are there any ideas on this subject, wishes, suggestions? I will be glad to know your opinion on this issue.
Thanks for attention.
Source: https://habr.com/ru/post/105898/
All Articles