Geographical reference and geometric correction of images. Rationale for the use of these procedures, examples of tasks. Implementation in the ERDAS Imagine package. Linking a raster map

We increase the extent of the base map to the first anchor point. We also enlarge the image being anchored to the selected anchor point. Having approached the anchor point in the module window, click the “Add point” button and click the mouse pointer on the selected point. After this, a form for entering coordinates opens. Coordinates can be entered either through input fields or captured from the map. If we have coordinates of points, for example, obtained using a GPS navigator, we can enter them in the appropriate fields. To obtain coordinates from the base map, click the “From Map” button.

The quality of the snap depends on the number of snap points and the choice of transformation method. You can read more about transformation methods.

If in the transformation parameters window you checked the “Open QGIS result” option, then after the binding process is completed, the resulting layer will be added to the working project on top of the base map.

An important nuance is that as a result of the module’s operation, the resulting raster has the coordinate system specified in the transformation parameters, but it does not contain information about exactly what projection of the raster it is. For this reason, it may be present in the list of layers, but not displayed on the map. In this case, you need to go to the “layer properties” and specify the right system coordinates manually.

More details

1. Matching images based on "features"

Literature for self-study

The book ($\textit(Krasovsky, Beloglazov, Chigin)$) contains a presentation of the classical theory of correlation-extremal analysis of two-dimensional fields, which we recommend that you familiarize yourself with it as part of an in-depth course.

An original approach to the mutual linking of images based on the so-called search-free correlation is outlined in the book ($\textit(Astapov, Vasiliev, Zalozhnev)$). This approach is more applicable in the field of correlation tracking than in the field of arbitrary image comparison, but it is attractive due to the possibility of efficient software and hardware-software implementation.

In the book ($\textit(Shapiro, Stockman)$), Chapter 11 is devoted to methods for comparing images and objects in two-dimensional space. The geometric aspects of the problem, which received less attention in our presentation, are of interest here. Chapters $12$ and $13$ cover the perception of 3D scenes. They can also be recommended for independent study, although the presentation of the same range of issues in the book seems to us more complete and successful.

In the book ($\textit(Forsyth, Pons)$) a small section “binocular image matching” is directly devoted to the problem of stereo identification, which at the same time contains a number of interesting ideas that are absent in our presentation. In particular, the stereo identification method is described dynamic programming and a number of other methods. In a broad sense, the entire part III of this book is devoted to the problem of reconstructing three-dimensional spatial information from a set of two-dimensional images, including chapters $10$ “Geometry of several projections”, $11$ “Stereovision”, $12$ “Determination of affine structure from motion” and $13$ “Determination of aprojective structures by movement." The issues discussed here are related to the construction of various metric and projective relationships between image points and scene points, calculation of ray paths, etc. These issues are not included in this training course, since they are closer to the photogrammetric field than to the field of image processing and analysis, however, within the framework of an advanced computer vision course, such elements should be considered necessary. In this regard, we recommend the entire III part of the book for in-depth independent study.

List of sources by section

1. $\textit(Bertram S.)$ The UNAMACE and the automatic photomapper\Dslash Photogrammetric Engineering. 35. No.6. 1969. P.569 - 576.
2. $\textit(Levine M.D., O"handley D.A., Yagi G.M.)$ Computer Determination of Depth Maps\Dslash Computer Graphics and Image Processing. 2. No.2. 1973. P.131 - 150.
3. $\textit(Mori K., Kidode M., Asada H.)$ An iterative prediction and correction method for automatic stereocomparison\Dslash Computer Graphics and Image Processing. 2. No.3 - 4. 1973. P.393 - 401.
4. $\textit(Ackerman F.)$ High precision digital image correlation\Dslash IPSUS. 1984. No. 9. P.231 - 243.
5. $\textit(Gruen A., Baltsavias E.)$ Adaptive least squares correlation with geometrical constraints\Dslash SPIE. 1985.V.595. P.72 - 82.
6. $\textit(Ohta Y., Kanade T.)$ Stereo by intra- and inter-scanline search using dynamic programming\Dslash IEEE PAMI. V.7. No.2. 1985. P.139 - 154.
7. $\textit(Priice K.E.)$ Relaxation techniques for matching\Dslash Minutes of the Workshop of Image Matching, September 9-11, 1987, Stuttgart University, F.R.Germany.
8. $\textit(Foerstner W.)$ A feature based correspondence algorithm for image matching. ISPRS Commision III Symposium, Rovaniemi, Finland, August 19-22, 1986\Dslash IAPRS. V.26-3/3. P.150 - 166.
9. $\textit(Ayache N., Faverjon B.)$ Efficient registration of stereo images by matching graph description of edge segments\Dslash IJCV. V.1. No.2. 1987. P.107 - 131.
10. $\textit(Van Trees G.)$ Theory of detection, estimation and modulation. T.1 - M.: Soviet radio, 1972.
11. $\textit(Vasilenko G.I., Tsibulkin L.M.)$ Holographic recognition devices. - M.: Radio and communication, 1985.
12. $\textit(Bochkarev A.M.)$. Correlation-extreme navigation systems\Dslash Foreign radio electronics. 1981. No. 9. P.28 - 53.
13. $\textit(Yaroslavsky L.P.)$ Digital processing signals in optics and holography: Introduction to digital optics. - M.: Radio and communication, 1987.
14. $\textit(Horn B.K.)$ Robot vision. - M.: Mir, 1989.
15. $\textit(Denisov D.A., Nizovkin V.A.)$ Segmentation of images on a computer\Dslash Foreign radio electronics, No. 10. 1985.
16. $\textit(Davies E.R.)$ Machine Vision: Theory, Algorithms, Practicalities. - Academic Press., 2nd Edition, San Diego, 1997.
17. $\textit(T. Tuytelaars, L. Van Gool.)$ Matching widely separated views based on affine invariant regions\Dslash International Journal of Computer Vision 59(1). 2004. P.61 - 85.
18. $\textit(Yaroslavsky L.P.)$ Accuracy and reliability of measuring the position of a two-dimensional object on a plane\Dslash Radio engineering and Electronics. 1972. No. 4.
19. $\textit(Abbasi-Dezfould M., Freeman T.G.)$ Stereo-Image Registration Based on Uniform Patches, International Archives of Photogrammetry and Remote Sensing. V. XXXI. Part B2. Vienna, 1996.
20. $\textit(Schenk.)$ Automatic Generation of DEM`s, Digital Photogrammetry: An Addentum to the Manual of Photogrammetry\Dslash American Society for Photogrammetry(\&)Remote Sensing. 1996. P.145 - 150.
21. $\textit(Gruen A,)$ Adaptive Least Squares Correlation: A powerful image matching technique\Dslash South African Journal of photogrammetry, Remote Sensing and Cartography. V.14. Part 3. June, 1985.
22. $\textit(Golub G.H., Ch. F. Van Loan.)$ Matrix computations. - John Hopkins University Press, 1983.
23. $\textit(Pytyev YP.)$ Morphological analysis of images\Dslash Reports of the USSR Academy of Sciences. 1983. T.269. No. 5. C.1061 - 1064.
24. $\textit(Haralick R.M. and Shapiro L.G.)$ Machine vision. - Addison-Wesley, 1991.
25. $\textit(Zuniga O.A., Haralick R.M.)$ Corner detection using the facet model\Dslash Proc. IEEE Comput. Vision Pattern Recogn. Conf., 1983. P.30-37.

The raster map in the GIS "Map 2000" is in RSW format. The format was developed in 1992, its structure is close to the TIFF version 6 format. The main indicators characterizing a raster map are:

• image scale;
• image resolution;
• image size;
• image palette;
• planned image linking.

Image scale- a value characterizing the source material (as a result of scanning which this raster image was obtained). Image scale is the relationship between the distance on the source material and the corresponding distance on the ground.

Image Resolution- a value characterizing the scanning device on which the raster image was obtained. The resolution value shows how many elementary dots (pixels) the scanning device divides a meter (inch) of the original image into. In other words, this value shows the size of the “grain” of the raster image. The higher the resolution, the smaller the “grain”, which means smaller size terrain objects that can be uniquely identified (deciphered)

Image Size(height and width) - values ​​that characterize the image itself. From these values ​​it is possible to determine dimensions raster image in pixels (points). The image size depends on the size of the source material being scanned and the resolution setting.

Image palette- a value characterizing the degree of display of color shades of the source material in a raster image. There are the following main palette types:

• two-color (black and white, one digit);
• 16 colors (or shades of gray, four digits);
• 256 colors (or shades of gray, eight digits);
• High Color (16 bits);
• True Color (24 or 32 bits).

If it is possible to select the resolution and image palette when scanning source materials (some scanning devices only work with fixed values), it should be taken into account that when increasing the resolution and choosing a higher degree of shade display, the volume of the resulting file increases sharply, which will subsequently affect the volume stored information and speed of display and processing of raster images. For example, when scanning source map materials, there is no need to use a palette of more than 256 colors, since in reality, as a rule, there are no more than 8 colors on a regular map.

The image palette is stored in source file, and the resolution and scale of the future image should be entered when converting the raster to an internal format. The exception is TIFF files, which store resolution in addition to the palette. For other cases, the resolution is indicated in accordance with the parameters selected during scanning. For example, domestic drum scanners from KSI scan with a resolution of 508 dots/inch (or 20,000 dots/meter). If you do not know the exact scale value of the processed materials, you should enter an approximate value (the scale value is automatically specified during the process of linking a raster image).

A raster image loaded into the system is not yet a raster map, since it does not have a planned reference. An untethered image is always added to the southwestern corner of the map dimensions. Therefore, if you are working with a large area of ​​work, to quickly search for an added raster, you can use the “Go to raster” item in the raster image properties menu of the “Raster List” dialog.

Once linked, the raster map becomes a measuring document. Using a raster map, you can determine the coordinates of the objects depicted on it (when you move the cursor along the raster map, the current coordinates are displayed in the information field at the bottom of the screen). A linked raster map can be used as a stand-alone document or in conjunction with other data.

1.2. Converting raster data

The Panorama system processes raster maps presented in RSW format (internal system format). Data from other formats (PCX, BMP, TIFF) can be converted to RSW format using software Panorama systems. In addition, the system supports early version raster data structures RST ("Panorama under MS-DOS"). When you open an RST file, it is automatically converted to RSW format.

There are two ways to load a bitmap into the system:

• Opening a raster image as an independent document (the "Open" item in the "File" menu).
• Adding a bitmap to an already open document(vector, raster, matrix or combined map). Adding a raster image to an already open map is done through the "Add - Raster" item in the "File" menu or the "Raster List" item in the "View" menu.

1.3. Linking a raster map

Binding raster map is carried out according to the linked document, i.e. First, you need to open a document oriented in a given coordinate system (vector, raster or matrix map), add the raster to be referenced and perform the reference. You can link a raster using one of the methods provided in the raster properties ("List of rasters - Properties"). It should be remembered that all raster actions available in the raster image properties menu are performed on the CURRENT raster. Therefore, if an open document contains several rasters, you should activate (select in the list of open rasters) the one with which you are in given time want to work.

1.3.1. Snap by one point

Snapping is done by sequentially indicating a point on the raster and the point where the specified point should move after the transformation (from where to where). The transformation is performed by moving the entire raster in parallel without changing its scale or orientation.

1.3.2. Move to southwest corner

The transformation is carried out by parallel movement of the entire raster without changing its scale and orientation to the southwestern corner of the dimensions of the work area. It is advisable to use this snapping mode when you add an incorrectly linked raster to an open map, which is displayed far outside the work area. In this case, after moving the raster to the southwest corner, it is easier to re-snap it.

1.3.3. Two-point snapping with scaling

The binding is done by sequentially specifying a pair of points on the raster and the points to which the specified points should move after the transformation (from where to where, from where to where). The transformation is performed by moving the entire raster in parallel and changing its scale. The image is snapped using the first pair of specified points. The second pair of points is specified to calculate the new scale of the raster image. Therefore, if the raster has unequal vertical and horizontal scale values ​​(the raster is elongated or compressed due to deformation of the source material or an error in the scanning device), the second point will take its theoretical position with some error. To eliminate the error, you should use one of the methods for transforming a raster image (application task "Transforming raster data").

1.3.4. Rotate without scaling

The binding is done by sequentially specifying a pair of points on the raster and the points to which the specified points should move after the transformation (from where to where, from where to where). The transformation is carried out by parallel movement of the entire raster with a change in its orientation in space. The rotation is carried out around the first specified point. The image is snapped using the first pair of specified points. The second pair of points is specified to calculate the image rotation angle. Therefore, if the raster has unequal vertical and horizontal scale values ​​(the raster is elongated or compressed due to deformation of the source material or an error in the scanning device), the second point will take its theoretical position with some error. To eliminate the error, you should use one of the methods for transforming a raster image (application task "Transforming raster data").

When loading raster maps into the database, a raster map work area can be created. To create a raster region, it is necessary to sequentially load into the system each raster image forming this region and orient it relative to unified system coordinates
The combination of raster and vector maps for the same or adjacent territories allows you to quickly create and update work areas, while maintaining the ability to solve applied problems for which some types of map objects must have a vector representation.

Graphic objects (drawings and images) located in Word document, as a rule, can be moved along with the text or tied to a specific fragment text document(paragraph, page boundaries, line, etc.).

To do this, enter the menu command FORMAT ® Drawing (Autoshape, Inscription or etc.) and in the corresponding dialog box on the tab Position click on the button Additionally and then open the tab Pattern position and set the switch Move with text. Typically, the mode for moving graphic objects along with text is set by default in Word.

To display the binding you need to enter the command SERVICE ® Parameters and on the tab View dialog box Options set switch Snap objects. When you install this switch after selection graphic object next to it (in the left margin) will be displayed anchor symbol (marker) in the form of an anchor.

Anchor symbols are displayed only in page (and Web document) layout mode and only for pictures and objects located outside the text layer(for which one of the modes is set text wrapping).

When working with a document containing a graphic object, it is recommended not only to set the display of anchor characters, but also to display non-printing characters (paragraph markers). Because when you delete, move or copy a paragraph near which an anchor symbol (anchor) is set, the graphic object (drawing or image) “anchored” to this paragraph is also deleted (moved, copied) along with the paragraph.

Sometimes you want a graphic to remain anchored to the same paragraph no matter how you move it, i.e. was “rigidly” tied to a specific fragment of the document, for example, a drawing to its title. In this case, in the Additional Markup dialog box on the Picture Position tab, you need to activate the switch Set binding, after which a castle image will be added to the anchor image in the anchor marker.

Creating formulas

Complex mathematical equations, expressions and formulas created using the built-in software can be inserted as objects into a Word document. Word editor formulas - programs Microsoft Equation.

The equations and formulas created in this way are static objects, i.e. they do not perform calculations and cannot be edited directly in the text.

To launch the formula editor, use the command Insert ® Object. In the dialog box that opens Inserting an object on the tab Creation select item Microsoft Equation 3.0. After this, the formula editor program menu and toolbar will appear on the screen Formula.

In addition, to launch the formula editor, you can use the button Formula editor.

When creating formulas, you use the formula editor toolbar buttons to select symbols and templates, and use the keyboard to enter numbers and variables in specially designated spaces.

The formula editor toolbar (Formula) contains two rows of buttons. In the top row - in the line characters there are buttons for inserting mathematical symbols into the formula - Greek letters, mathematical and logical operators, superscripts, etc. The bottom row buttons allow you to insert templates , including symbols for fractions, square roots, integrals, sums, products, matrices, various parentheses, etc. Many templates contain special fields (black or empty squares) for entering text and inserting characters.

Entering and editing formulas is completed by pressing the ESC key or closing the formula editor panel. You can also left-click anywhere in a document field outside the formula entry area. The entered formula is automatically inserted into the text as an object. Then it can be moved to any other place in the document via the clipboard. To edit a formula directly in the document, just execute double click. This automatically opens the formula editor window.

Creating tables and working with tables in Word

Word allows you to format the data of created documents in the form of tables.

Table– a form of organizing data into columns and rows, at the intersection of which there are cells. Table cells can contain data of any type: text, numbers, graphics, pictures, formulas, etc.

A Word table can contain 63 columns and 32,767 rows (compare Excel - 256 columns and 65,536 rows). Different rows of the same table can contain different numbers of columns. Table cells have addresses formed by the column name (A, B, C,...) and row number (1,2 3,...).

In a Word document, tables are created at the location of the cursor. By default, lines in the table are indicated by dotted lines (which are not printed).

You can create a new table in Word format:

1. Using the window's horizontal menu command TABLE ® Add (Insert) ® Table. In the dialog box that appears Inserting a table you should set the table dimension - the number of rows and columns and set the column width parameters.

2. Using the Add table button on the standard toolbar. To define the configuration of a new table, you need to color in the required number of columns and rows of the table while holding down the left mouse button.

3. B latest versions Word now has the ability to create tables by drawing them with a “pencil” using the mouse. This button is located on the toolbar Tables and borders.

4. Previously typed text can be converted into a table view using the command TABLE ® Convert ® Convert to Table provided that the text is prepared using special line and column separators: end-of-paragraph characters ( Enter), tabs ( Tab) or others.

Word also allows you to convert a table back to plain text using the menu command TABLE ® Convert ® Convert to text.

The number of rows and columns initially specified (when creating a Word table) can be changed by adding new rows and columns or deleting existing ones.

For adding new line at the end of the table you need to place the cursor in the last cell of the table and press the key Tab.

You can also use the clipboard to move, copy, add, and delete individual table cells, columns, and rows (menu commands EDIT ® Copy, Cut, Paste).

To delete a table, you need to select it along with paragraph marker, next to the table, and press the key Delete. If you select a table without a paragraph marker following the table, pressing a key will delete only its contents. You can also use the command to delete the entire table TABLE ® Delete ® Table, having previously positioned the cursor inside the table.

New features for working with tables in Word 2000

In the version of Word 2000, for the convenience of working with tables, new tools and capabilities appeared that were not available in previous versions Word:

· moving the entire table with the mouse - drag the table movement marker with the mouse - a non-printing symbol that appears on the left above top line tables;

· changing the size of the table while maintaining the proportions of the sizes of rows and columns (using the table resizing marker in the lower right corner of the table);

· text wrapping around the table (the wrapping options are set in the same way as for pictures - command TABLE ® Table Properties);

· creating nested tables – a table cell can contain another table (for example, using the command TABLE ® Add ® Table);

· creating diagonal borders and lines inside a cell by drawing borders with a pencil or using buttons on the toolbar External boundaries;

· setting cell margins and intervals between cells, etc. (cell fields determine the gap between the cell border and the text in it; to set cell margins and determine the amount of interval between cells, use the command TABLE ® Table Properties ® Table tab ® Options button).

A. P. Kirpichnikov, D. I. Miftakhutdinov, I. S. Rizaev

SOLUTION OF THE PROBLEM OF CORRELATION OF IMAGE AND DIGITAL MAP OF THE TERRITORY

Keywords: image combination, digital terrain map, correlation image processing.

The paper discusses the solution to the problem of linking an image and a digital map of the area using the method of correlation processing of two images, which makes it possible to achieve high accuracy of the link to automatically eliminate alignment errors between them.

Keywords: combining images, digital terrain maps, correlation image processing.

The work considers the solution of the binding images and digital maps by method of correlation processing of the two images to achieve high accuracy of snapping for automatically eliminating alignment errors between them.

Introduction

Currently in Russian existing systems reconnaissance, the main goal is to find new (previously unknown) objects in a given area of ​​the terrain. Therefore, an important task is to combine the terrain map (DTM) and its current image with subsequent analysis of the results of the combination and search for differences.

In practice, multi-temporal and multi-spectral images of the same object or area can differ significantly from each other and from their image on the digital digital computer. Thus, we are faced with a number of tasks of geometric and amplitude correction of images, their alignment and alignment. It is possible to bind using navigation parameters and using search algorithms, establishing correspondence between image elements.

Errors in the measurement of navigation parameters lead to errors in the alignment of the image and the DCM. The main reasons are:

1. Delay in the start of signal reception during image formation.

The error in determining the delay is formed due to the discreteness of the value clock frequency reference oscillator (e.g. 1/56 MHz)

56 10 6 [Hz] 2 56 10

2. Error in determining media height. Numerical error calculation (approximate):

3. Error in determining the boundaries of the image frame.

This error is determined primarily by the error of the angle sensor. The maximum linear error due to the error is determined

as Dmax STr = 1.74-10-3 Dmax.

4. Error in determining the coordinates of the aircraft in the ground coordinate system.

where D is the range to the point in the image frame, h is the height of the aircraft, D is the error in measuring the height of the aircraft, D is the error in determining the angular position of the antenna in radians, D^ is the error in determining the true heading of the aircraft in radians.

The total error in determining the image location is equal to the square root of the sum of the squares of the component errors.

To eliminate the resulting registration errors, it is possible to use correlation binding of processed images and DCM. At the same time, the main difficulties in creating algorithms include, first of all, differences in the principles of image acquisition. In addition, the images of most objects depend significantly on the time of year. Therefore, when creating an algorithm for correlating images and DCM, it is necessary to be able to identify landmarks with stable characteristics.

Basic concepts of correlation and regression analysis

The main task of correlation analysis is to estimate the regression equation and determine the closeness of the relationship between the resulting characteristic and a variety of factor characteristics. The value of the correlation coefficient is an expression of the quantitative closeness of the connection.

If we consider the general population, then to characterize the closeness of the relationship between two variables, we use the pair correlation coefficient p, otherwise, its assessment is the sample pair coefficient r.

If the form of the relationship is linear, then the pair correlation coefficient is calculated using the formula:

and the sample value - according to the formula:

Y(X - X)(Y -Y)

With a small number of observations, the sample correlation coefficient is calculated using the formula:

pX X T-X XX T

X X,2 - (X X)2

"X t 2 - (X T)2

Changes in the value of the correlation coefficient are in the range -1< г < 1.

If the correlation coefficient is in the range -1< г < 0, то между величинами Х и У - обратная корреляционная связь. Если коэффициент корреляции находится в интервале 0 < г < 1, то между величинами Х и У - прямая корреляционная связь.

Logic for applying correlation binding

The main stages when combining include:

1. Identification of standards from the map, their pre-processing.

2. Transformation of image standards taking into account the geometry of the resulting image.

3. Image processing to highlight terrain objects.

4. Carrying out a correlation search for standards in the current image.

5. Clarification of the position of the combined image with the map (correction of navigation coordinates).

Let's take a closer look at some of the stages.

Obtaining standards

This stage is carried out by the operator or automatically based on knowledge of the intended area of ​​action and the objects located on it, which can be divided into two groups. The first is point-based, in particular - towers, structures, etc. To highlight them in the image, you can use thresholding of image brightness values. However, the main difficulty arises when associating a given “bright” point with a terrain object, due to the fact that the threshold can be exceeded by another object. Insufficient detail of digital maps does not allow, in most cases, to identify point objects on the ground.

The second group includes extended objects with characteristic shapes. These include hydrography (rivers, lakes, coastline), road network, settlements etc. These objects have characteristic images and allow, based on knowledge of their properties on the map, to obtain an image model for subsequent search. Research has shown the advisability of reducing standards to a binary form due to the fact that it is impossible to predict the brightness level of objects in the generated images. Figure 1 shows the acquisition of a binary image of a river using a DCM.

Rice. 1 - An example of obtaining a binary image of a river using a DCM

It is advisable to select characteristic areas of objects as reference ones, such as bends, intersections, and branches. They have narrow autocorrelation functions and should provide efficient search. It is possible to use an automatic algorithm for selecting the position of reference sections by analyzing the correlation function of the selected sections and the area from which they are formed. The landmarks used are selected for the intended area of ​​the terrain, obtained from the readings navigation system taking into account the possible magnitude of its error.

Elimination of geometric distortions

An issue that requires consideration when implementing the correlation binding algorithm is the choice of the transformed area. In this case, two options are possible. The first is bringing reference terrain areas to the current image. This operation is more advantageous from the point of view of computational resources, since it is easier to process the binary reference image. The second method involves bringing the current image to a map of the area. The choice of transformation method is carried out taking into account the possibilities of direct implementation of the algorithms in practice.

Processing of received images

Directly searching for reference areas in the resulting images is impractical due to large quantity objects on the ground, the presence of a significant noise component. Therefore, the search stage is preceded by the operation of selecting the desired objects. The main methods currently used to perform this operation are image segmentation and contouring. In addition, to reduce the dependence of image processing results on distorting random noise components, image filtering is performed. In this case, certain components of the image itself can act as interference.

Segmentation is often considered as the main initial stage of analysis when automating image acquisition methods, since the result is an image, the quality of which largely determines the success of solving the problem of identifying objects in the image and further correlation. Example of threshold binary

The segmentation of the resulting and transformed image is shown in Fig. 2.

Fig.2 - Example of a transformed image

Please note that to highlight various objects must be implemented different ways image processing. Thus, to highlight straight sections of roads, you can use special masks followed by threshold processing.

Finding the location of reference images on the current image (snapping)

The main variants of algorithms for establishing image similarity are associated with obtaining characteristics of the stochastic relationship of the current image fragment with a reference image of the area. The basis of these algorithms is the correlation and spectral theory of signals.

The image of the reference fragment (selected on the terrain map and represented by a matrix u0 of size pxn) is compared with the current images by image fragments in the “zone of interest” of size bxb. b=n+m, and the search area is determined possible mistake navigation systems.

During the sliding search process, a “similarity function” is calculated between fragments of the reference and current images. It is necessary to find a similarity function that, with maximum accuracy and reliability, will allow you to localize a fragment of the image corresponding to the standard, thus establishing conjugate points in the images.

With the correlation method, a search is made for the maximum correlation coefficient (max(k,1)) of the current fragment with the standard

XXUo(x, Y)u(x, y)

/(k, I) =-^-]-_, (7)

^[^x, y)]2 XX2)2

where u0 and u are the centered brightness values ​​of the standard and the image fragment. This operation is necessary to eliminate the dependence of the correlation coefficient value on the energy of the areas.

To comply with the detection reliability conditions, it is necessary to set a threshold (gthor) for the cross-correlation value.

If max(k,1)>gpor, then the similarity of the found pair of fragments is guaranteed with a given probability.

Comrade The threshold value can be set by the probability of similarity of fragments and the correlation coefficient.

The disadvantage of the correlation similarity measure is its sensitivity to geometric distortions in the sizes of mating objects, which places high demands on the object segmentation algorithm based on the resulting image.

Usually, the accuracy of fragment combination and the probability of false binding are taken as criteria for the effectiveness of similarity identification procedures.

Figure 3 shows the search results for several reference fragments per image. The standards identified on the DCM are reduced to the geometry of the resulting image. Figure 4 shows the result of searching for a reference image in the case of reducing the image to the map geometry under the same conditions.

The relationship between the reference and the image can be calculated based on the spectral theory of signals. In fact, the method also searches for the correlation integral, only in the frequency domain. In this case, using fast Fourier transform algorithms, it is possible to significantly reduce the required computational costs for organizing calculations.

Based on the obtained values ​​of the discrepancies between the predicted result of navigation and the positions of the reference calculated using the correlation integral, a correction to the position of the current image relative to the DCM is formed.

Rice. 3 - Search results for several reference fragments

Rice. 4 - Result of searching for a reference image in the case of reducing the image to the map geometry

The considered method of correlation processing of two images allows us to achieve high accuracy in linking the current image with a digital terrain map to automatically eliminate registration errors between them.

The paper proposes an algorithm for performing the binding, the main stages of which are the preparation of standards from the map, transformation and processing of terrain images and the implementation of correlation search. However, each of these stages during implementation requires taking into account the features of the survey systems used and digital maps of the area.

Literature

1. Baklitsky V.K. Correlation-extreme methods of navigation and guidance / Tver Publishing House: TO “Book Club”, 2009. - 360 p.

2. Gruzman I.S., Kirichuk V.S., Kosykh V.P. etc. Digital image processing in information systems./ Tutorial. - Novosibirsk: NSTU Publishing House, 2000. -168 p.

3. Kirpichnikov A.P., Miftakhutdinov D.I., Rizaev I.S. Solving the geopositioning problem using the correlation comparison method // Bulletin of the Technological University: T.18 No. 3; - 2015. - 308 p.

4. Miftakhutdinov D.I., Rizaev I.S. Features of the implementation of algorithms for combining images with digital terrain maps./ “Prospects for the integration of science and practice.” Materials of the II International Scientific and Practical Conference; Stavropol: 2015. - 94 p.

© A. P. Kirpichnikov - Doctor of Physics and Mathematics. sciences, head department intelligent systems and control information resources BOOK, [email protected]; D. I. Miftakhutdinov - 2nd year master's student of the department automated systems information processing and management KNIGU-KAI; [email protected]; I. S. Rizaev - Ph.D. those. Sciences, Professor of the Department of Automated Information Processing and Control Systems of KNIGU-KAI; [email protected].

© A. P. Kirpichnikov - Dr. Sci., Head of the Department of Intelligent Systems & Information Systems Control, KNRTU, [email protected]; D. I. Miftakhutdinov - master student of the Department of Automated information processing and management, KNRTU-KAI, [email protected]; I. S. Rizaev - PhD, Professor of the Department of Automated information processing and management, KNRTU-KAI, [email protected].