RU2266565C1 - Recognizing device - Google Patents

Abstract

FIELD: computer science, possible use for automatic detection and recognition of images in objects in cybernetics, geology, biology, physics, medicine, etc.

SUBSTANCE: known recognizing device additionally has coherent-optical processor unit, allowing instant detection and marking of suspicious fragment with target object (set of signs) at input image, and determining of its form, scale and coordinates.

EFFECT: decreased amount of necessary calculations, decreased recognition time, higher speed of operation of device.

2 dwg

Description

The invention relates to the field of computer technology and can be used for automatic detection and recognition of images and objects in robotics, geology, biology, physics, medicine, etc.

The problem of pattern recognition (images) continues to cause great interest of engineers and researchers of various specialties: radio engineers, physicists, geologists, biologists, physicians, developers of industrial robots, specialists in space research, etc. The most relevant and complex are rightfully considered recognition tasks associated with the analysis of video images. This is explained, in particular, by the convenience of presenting various kinds of information in the form of images and the high throughput of a two-dimensional video channel.

The most common approach to pattern recognition is to establish a correspondence between the current (input) image and the reference image of the object to be detected (recognized).

A recognition device is known (US Patent No. 5061063 dated October 29, 1991 [L.1]) containing an acousto-optic correlator (processor), threshold devices, and execution units. This device recognizes simple images by comparing them with predefined (reference) images. The disadvantage of this device is the high probability of false alarms, especially when recognizing complex objects that make up part of the input image.

A recognition device is known (US Patent No. 6519352 dated February 11, 2003 [L.2]), comprising an electronic computer with electrically connected in series: an input television camera, a video buffer, an image pickup unit and an electronic processor, as well as an electrically connected feature memory unit objects (filters) and a filter sampling unit, the output of which is connected to the second input of an electronic processor that authenticates the signal (recognition) according to the algorithm specified by the filter. Information hidden in the image (signal) can be extracted even after changing the propagated signal by the compression algorithm with data loss. Authentication can only be carried out by the person holding the key - a set of signs and rules by which the image (signal) is identified. Authentication requires search programs whose algorithms are not considered in the patent.

The disadvantage of this device is its low performance, since it is necessary to search for a zone with known (identifiable) features in a huge volume of the entire input signal (image). Even worse, when the search for known (identifiable) signs must be performed in an unlimited amount of input information (for example, among current television images or many photographs in all illustrated magazines). The situation becomes completely disastrous when there is a huge number of different sets (groups) of signs, for each of which a separate identification of various images (signals) is made.

The essence of the invention lies in the fact that in a recognition device containing an electronic computer with electrically connected in series: an input television camera, a video buffer, an image pickup unit and an electronic processor, as well as an electrically connected object memory (filters) memory unit and a filter sampling unit, output which is connected to the second input of an electronic processor that performs recognition by any known algorithm, the following are introduced: a coherent optical processor with a block quickly changing filters and a CCD camera, as well as a multiplexer, a threshold block, a block for determining the filter number, a block for determining coordinates X and Y and a block for forming windows X ± ΔX and Y ± ΔY, while the electrical output of the block for quickly changing filters is connected via a block for determining filter numbers (filter register) with the first input of the multiplexer and the control input of the filter selection block of the electronic computer filter. The first output of the CCD camera through the unit for determining the coordinates of X and Y (coordinate register) is connected to the second input of the multiplexer and through the unit for forming windows X ± ΔX and Y ± ΔY with the control input of the image pickup unit of the electronic computer. The second output of the CCD camera through the threshold block is connected to the control inputs of the filter number determination unit and the X and Y coordinate determination unit. The output of the electronic computer is connected to the control input of the multiplexer, and the output of the input television camera is additionally connected to the input of the coherent optical processor. In addition, the output of the filter number determination unit is additionally connected to a control input of the window forming unit X ± ΔX and Y ± ΔY.

With such a recognition device, the coherent-optical processor allows you to instantly identify suspicious places in the current image and tell the electronic computer where the image needs to be recognized and which algorithm (filter) you need to use for this purpose in the memory. Such an indication dramatically reduces the number of computational operations required for recognition of an electronic computer and speeds up the recognition process. The introduction of an additional connection between the filter number determining unit and the window forming unit X ± ΔX and Y ± ΔY allows you to adapt the size of the image (window) to be processed by the electronic computer with the size of the suspicious area detected by the coherent optical processor, which also reduces the number of recognition of computing operations of an electronic computer and speeds up the recognition process.

The aforesaid is illustrated by drawings, where it is represented:

figure 1 is a block diagram of a recognition device;

figure 2 is a simplified functional diagram of a recognition device.

The block diagram of a recognition device (figure 1) shows:

1 - input television camera, for example WAT902H (company WATEC CO. LTD, Japan);

2 - an electronic computer, for example, the GigaNT model 2600/121 series (POLARIS, Russia), with the VIVO video buffer, [L.3];

3 - video buffer of an electronic computer (SVGA Gigabite ATI Radeon 9200 128 Mb VIVO), [L.3];

4 - block image sampling of an electronic computer;

5 - a block of memory features of objects (recognition algorithms) of an electronic computer;

6 - block sample filters (recognition algorithms) of an electronic computer;

7 - an electronic processor (final classifier) of an electronic computer;

8 - coherent optical processor, for example, the company Lytton Data Systems [L.4];

9 - laser;

10 - beam former;

11 - space-time light modulator (PVMS) with a driver;

12 - lens direct Fourier transform;

13 is a block of rapidly changing filters (filter memory block) of a coherent optical processor;

14 - lens inverse Fourier transform;

15 - CCD camera of a coherent optical processor with a driver, for example, a CA-D6-256W MotionVision camera (from DALSA INC. Canada) with a PC-DIG network card (from Coreco Imaging, Inc., USA);

16 - threshold block;

17 - block determining the filter number (filter register);

18 - block determining the coordinates of X and Y (register of coordinates);

19 is a block forming windows X ± ΔX and Y ± ΔY;

20 - multiplexer.

Blocks: 9 (laser), 10 (beam former), 11 (spatio-temporal light modulator), 12 (direct Fourier transform lens), 13 (filter block of coherent-optical processor filters), 14 (reverse Fourier transform lens), 15 (CCD camera) - are elements of a coherent optical processor 8, for example, the firm Lytton Data Systems [L.4];

blocks: 3 (video buffer of an electronic computer), 4 (image selection block), 5 (object feature memory), 6 (filter selection block), 7 (electronic processor), 16 (threshold block), 17 (number determination block filter), 18 (unit for determining the coordinates of X and Y), 19 (unit for forming windows X ± ΔX and Y ± ΔY), 20 (multiplexer) - are elements of an electronic computer 2 (for example, a personal computer of the GigaNT series model 2600/121 [L.3]).

The recognition device includes an input television camera 1, the output of which is electrically connected to the input of the video buffer 3 of the electronic computer 2 and the input of the coherent optical processor 8. The optical output of the coherent optical processor is connected to the input of the CCD camera 15, the electrical output of which is connected to the threshold device 16 The output of the threshold device 16 is electrically connected to the control inputs of the filter number determination unit (filter register) 17 and the X and Y coordinate determination unit 18. Synchronizing in the moves of the CCD camera 15 (clock, line and frame) are connected to the inputs of the X and Y 18 coordinate determination unit, the output of which is connected to the first input of the multiplexer 20 and the input of the window forming unit X ± ΔX and Y ± ΔY 19. The output of the window forming unit X ± ΔX and Y ± ΔY 19 is connected to the control input of the image pickup block 4 of the electronic computer 2. The electrical output of the set of rapidly changing filters 13 of the coherent optical processor 8 is connected to the input of the filter number determination unit (filter register) 17, the output of which is connected to the the direct input of the multiplexer 20, the control input of the filter sampling unit 6 of the electronic computer 2 and the control input of the window forming unit X ± ΔX and Y ± ΔY 19. The output of the video buffer 3 of the electronic computer 2 is connected to the image sampling unit 4, the output of which is connected to the first input electronic processor 7. The output of the feature memory block (feature set or recognition algorithm) of objects 5 of the electronic computer 2 is connected to a filter sampling unit 6, the output of which is connected to the second input a processor 7. Processor 7 is connected to output the third input of the multiplexer 20.

The device operates as follows. The input image from the television camera 1 enters the video buffer 3 of the electronic computer 2 and, at the same time, to the space-time light modulator 11 in the coherent optical processor 8. In the video buffer 3 of the electronic computer 2, the input image is delayed while it is processed in the coherent -optical processor 8. The basis of the coherent-optical processor 8 is a coherent optical correlator, consisting of optically sequentially connected laser 9, beam former 10, spatially a belt light modulator 11, a direct Fourier transform lens 12, a fast-changing filter unit 13 and an inverse Fourier transform lens 14, which allows instantly (at the speed of light) to obtain a correlation function between the input and reference images. In the coherent optical correlator, the input image signal supplied to the spatio-temporal light modulator (PVMS) 11 modulates the coherent radiation of the laser 9, expanded to the PVMS 11 image size by the beam shaper 10, and is converted into the spectrum by a direct Fourier transform lens 12. The resulting image spectrum multiplied in the block of rapidly changing filters 13 with the spectra of all reference images available in the bank of the filter block of the coherent optical processor 8 and after conversion by the lens about the Fourier transform 14 forms on the photosensitive matrix of the CCD camera 15 an image of the correlation field (a two-dimensional correlation function of the input image and the reference current currently in the block of rapidly changing filters). The output correlation field of the correlator is read by the CCD camera 15 of the coherent-optical processor 8 and converted into an output electric signal supplied to the threshold unit 16. The filter unit 13 of the coherent-optical processor 8 constantly issues 17 numbers to the inputs of the filter number determination unit (filter register) data) of the filter currently being processed by processor 8, and the CCD camera 15 (more precisely, the driver of the CCD camera 15 that reads the correlation field) issues the number of the read line to the X and Y 18 coordinate determination unit and reads aemogo currently pixel. If the specified threshold is exceeded by the output signal in any of the pixels read by the CCD camera 15, the threshold device 16 issues a command to the filter number determination unit (filter register) 17 and the X and Y coordinate determination unit 18. The filter number determination unit 17 remembers filter numbers, the value of the correlation function of which in the coherent optical processor exceeded the detection threshold specified in the threshold block 16, and transmits them (numbers) to the filter sampling unit 6 of the computing machine 2, which allows electronic computing Yelnia machine 2 to select from its memory 5 signs and recognition algorithm only the object, which was discovered by comparing the correlation in coherent optical processor 8, and not to touch features and algorithms for recognition of objects to be recognized. The unit for determining the coordinates X and Y 18 transfers the stored data to the unit for forming the windows X ± ΔX and Y ± ΔY 19. In the unit for forming the windows X ± ΔX and Y ± ΔY 19, the coordinates X and Y are shifted by the image size of the detected object ± x and ± y, for which the output of the filter number determination unit (filter register) 17 is additionally connected to the control input of the window forming unit X ± ΔX and Y ± ΔY. The output signal of the window forming unit is sent to the image sampling unit of the electronic computer, where the pieces corresponding to the generated windows are cut out from the full image stored by the video buffer 3, and they are processed by the final classifier (processor) 7 of the electronic computer according to the signs and algorithms selected by the sample block filter 6 for a given part of the input image. The electronic processor (final classifier) 7, deciding on the presence of the desired object in the input television image, issues a command to the multiplexer 20, allowing the latter to generate a work results protocol containing information about the coordinates of the recognized object in the input frame at the output of the recognition device (according to the data of the definition block coordinates X and Y 18) and that the object was recognized (according to the unit for determining the filter number 17). Thus, the presence of a coherent optical processor (correlator) in the proposed recognition device allows you to quickly determine which of all the desired objects is on the input television frame and its coordinates in this frame, and the electronic computer allows recognition (final classification) by pre-limited features in a sharply limited amount of input information (in a known place of the input frame, limited by the window size ± ΔX by ± ΔY), which provides yes Nome device positive effect: high performance and low probability of false alarms.

On a simplified functional diagram of figure 2 further explains the operation of the recognition device. As can be seen from the simplified diagram, the input frame from the television camera enters the correlator of the coherent-optical processor, operating at a speed of about 10 ¹⁴ op./s. Such correlator performance allows for one frame to receive correlation responses with thousands of filter standards and to reveal both the presence of the target and its coordinates (in the frame), as well as its image and image size in the frame. To do this, in the object signal received from the correlator (correlation response), a target is searched (threshold device 16 of FIG. 1) and its coordinates are determined (block 18 of FIG. 1). The type and size of the target image in the frame is determined by the data of the filter driver at the time of detecting the correlation response (block 17 of figure 1). Based on these data and coordinates, the electronic computer selects the fragment in which the object is detected, and also performs a detailed analysis of the fragment, taking into account the image and scale of the detected object. If, as a result of this analysis, the final classifier (electronic processor) makes a decision about the presence of an object, then a command is issued to the multiplexer to generate and issue a results protocol containing data on the recognized object.

The process of recognizing or determining the authenticity of a suspicious image can be described as follows:

1. Using the input camera (or scanner, thermal imager, locator) we get a suspicious signal or image.

2. If the image is not digitized, digitize.

3. We remember the image (signal) in the memory of an electronic computer (personal computer).

4. Using the holographic correlator of the coherent optical processor, we obtain a two-dimensional correlation field (response) of this image with each of the holographic filter standards available in the filter bank of the holographic correlator.

5. Using a CCD camera of a coherent optical processor, we convert each correlation field (response) of this image with each of the holographic filter standards available in the filter bank of the holographic correlator into an electric signal of a television type (line-frame structure).

6. If the image is not digitized, digitize.

7. Compare the signal of each correlation field with a threshold preselected by the criterion of false alarms.

8. If there is a response in the correlation field that exceeds a threshold, we determine:

- Which filter in the filter bank of the holographic correlator belongs to this correlation response.

- Coordinates of the peak of the correlation response exceeding the threshold in the correlation field.

9. The filter determines:

- Algorithm (filter) for further processing of the signal by an electronic computer (personal computer).

- The size of the image area to be further processed.

10. According to the coordinates of the peak of the correlation response that exceeds the threshold in the correlation field, we determine:

- The coordinates of the image area to be further processed.

11. From the image (signal) stored in the memory of the electronic computer (personal computer), we select a fragment whose size is determined by the filter and the position by the coordinates of the peak of the correlation response.

12. The selected image fragment is processed in an electronic computer (personal computer) with the previously defined algorithm (filter) for the final classifier (processor), for example, when encoding the image using the algorithm described in the prototype [L.2]:

- Subtract the selected suspicious signal from the original, receiving a difference signal.

- Consistently we receive correlations of a difference signal and all our random signals generated and recorded in the image before its distribution.

- Find the levels of zero and one using the sequence 0101 in the first four bits of the identification word. The mean of the first and third correlations will be level 0, and the mean of the second and fourth will be the unit level.

- We assign to each correlation result 0 or 1, using the found zero and one levels.

- Compare the result with the identification number.

- We will send a letter demanding compensation.

Literature:

1. US patent, MKI G 06 K 9/74; NCI 3 56/71; No. 5061063 of October 29, 1991 - an analogue.

2. US patent, MKI N 04 K 001/00; NKI 382/100; 380/252; No. 6519352 of February 11, 2003 - a prototype.

3. www.polaris.ru

4. Aviation Week & Space Technology, 12 / X 1998, p.50, 51.

Claims (1)

A recognition device containing an input television camera and an electronic computer with an electrically connected video buffer, an image pickup unit and an electronic processor, as well as an electrically connected object memory (filters) and a filter sampling unit, the output of which is connected to the second input of the electronic processor, the fact that a coherent optical processor with a block of rapidly changing filters and a CCD camera, as well as a threshold multiplexer, are introduced into it a block, a block for determining the filter number, a block for determining the coordinates of X and Y and a block for forming windows X ± Δx and Y ± ΔY, while the electrical output of the block for quickly changing filters is connected through the block for determining the filter number to the first input of the multiplexer controlling the input of the block for forming windows X ± Δx and Y ± ΔY and the control input of the filter sampling block of the electronic computer, the first CCD camera output is connected to the second input of the multiplexer through the X and Y coordinate determination unit and through the window formation block X ± Δx and Y ± ΔY with the control the input of the image sampling unit of the electronic computer, the second output of the CCD camera through the threshold block is connected to the control inputs of the filter number determination unit and the X and Y coordinate determination unit, the output of the electronic computer is connected to the control input of the multiplexer, and the output of the input television camera is additionally connected to input of a coherent optical processor.

Images