JP2009176163A - Section data creation system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To reduce the load and cost, when creating a farmland section map. <P>SOLUTION: Image data prepared beforehand is input, and edge data is extracted. Section data is obtained from the edge data. Section adjustment processing is performed and the section data is updated. Furthermore, part of section certainty degree data is acquired. Finally, the section data and the section certainty degree data are input, and section certainty degree determination processing is performed, for updating the section data and the section certainty degree data. In addition to the data in which graphical elements of the certainty degree are mainly input, colors and texture information are analyzed and detailed. By taking into consideration the arrangement in relation to surrounding sections, the form is partially corrected so that more natural data as a farmland map can be created. The section data and the section certainty data are the data targeted by this system, in which in addition to the simple and easy-to-use data of the farmland section, the value for self-determination as a guide, when checking them, is indicated as the certainty degree. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to a section data creation system, and in particular, using an image analysis system that analyzes a photographed image obtained by photographing a ground surface from an altitude (the sky) using a satellite, an aircraft, or the like. The present invention relates to a land parcel data creation system that extracts land boundaries and analyzes land plots that have been prepared mainly to create parcel data for those districts.

  In recent years, the use of satellite images, aerial photographs, and the like has become widespread, and in particular, it is actively used to efficiently search a large area in map creation and update processing.

  The advantage of using satellite images and aerial photographs is that the cost of surveying in the field as in the past can be omitted, and a wide range of data can be obtained at one time. Furthermore, it is possible to always prepare new data by increasing the imaging update frequency.

  In the past, the work of creating and updating a map using satellite images has been performed by a method in which the features and boundary lines are confirmed by visually observing the images, and these are manually input. The map creation work by visual inspection has a problem that the burden on the worker is large and the progress of the work and the accuracy of the result vary greatly depending on the skill level of the worker. At present, various methods for analyzing images using a computer have been studied and proposed for stable results and cost reduction.

  Generally, when analyzing an image, the first phase for extracting features such as shapes, patterns, and borders from the image, and the features acquired in the first phase are analyzed and processed into necessary information. To the second phase. The analysis result of the image by such a method is greatly influenced by the type and accuracy of the features extracted in the first phase of the first half. For this reason, it is important to appropriately extract features from the image.

  When creating a map from satellite images or aerial photographs, the various boundary lines found on the surface of the earth are particularly important features. These boundary lines are often detected based on edges (contour lines) of buildings, roads, and water areas. Many studies have been made on methods for extracting edges from images, and various practical proposals have also been made (for example, see Non-Patent Documents 1 and 2 below).

  As an example of a technique for detecting boundary lines such as roads and rivers from an image, analyzing the connection of the boundary lines, and automatically extracting land blocks, in Patent Document 1 below, an edge is first extracted from an image of the ground surface. There has been proposed a process of extracting polygonal areas formed by connecting a plurality of intersection points by extracting and obtaining intersection points between straight lines and edges extending radially from an arbitrary point on the edge image.

  Furthermore, in Patent Document 2 below, by extracting edges with different characteristics in multiple stages, it is possible to accurately extract the path in the farmland, and to easily create data that has been difficult to create, such as farm district image data. A system that can be created is proposed.

JP 2003-256807 A Japanese Patent Application No. 2007-70891 J. Canny, `` A Computational Approach to Edge Detection '' IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol.PAMI-8, No6, November 1986 Mamoru Onoe, “Image Processing Handbook”, Shoshoku Shuppan, 1987

  As a result of the above research, it has become possible to efficiently create and update maps using automatic recognition technology from satellite images and aerial photographs. However, due to changes in shooting conditions, noise included in the image, and features that do not apply to general features among the features to be recognized, it is practical to set the correct answer rate for automatic recognition to 100%. Impossible. Therefore, in order to create a highly reliable map, it is necessary to perform a visual check after automatic recognition by a computer, and then complement and correct necessary portions.

  Since this visual check is more expensive than automatic recognition by a computer, it is an important requirement for cost reduction to reduce the amount of work as much as possible.

  In the conventional method, automatic recognition by a computer outputs a result that is as faithful as possible to an image, and then visual check and correction are performed. However, when actually using a map, if it is desirable that the data is simple, there is a problem that the number of correction points increases. For example, when making a farm district map, if you try to give a recognition result faithfully to the image, the edges are rarely completely straight, and each section is often distorted, In actual use as a map, it is desirable that the rectangles are arranged as neatly as possible, so correction work is often required.

  In addition, there is no indication of whether or not the recognition result should be automatically checked, and eventually there is a problem that most data must be visually checked. For this reason, the cost of the check is large with respect to the correction amount, and whether or not the location requiring correction is noticed is often greatly influenced by the skill level of the operator.

  The present invention has been made in view of such a situation, and when correcting a feature extracted from a geographic image such as a satellite image or an aerial photograph image so as to be easily used in an actual map, each feature is provided. The object is to reduce the processing load and the cost by predicting the recognition accuracy.

  In order to deal with the above-mentioned problem, the present invention sets a value of “confidence” for the feature at the stage of extracting the feature from the image, and step by step how much it matches the feature to be extracted as the value. To express. Furthermore, when creating agricultural district image data, it is necessary to calculate the “confidence” of the section itself from the “confidence” of each feature that composes it, and to check it carefully when visually checking it. Display as an indication of whether or not there is. Also, when automatically creating agricultural district image data, the emphasis is not on the results faithful to the image data, but on the form in which the rectangles are neatly arranged so that there is no need for final correction. At this time, by using the certainty factor in the automatic shaping / correction process, the result of the correction can be made without distorting the recognition result as much as possible.

  According to one aspect of the present invention, there is a partition data creation system that analyzes geographic image data obtained by photographing from a high sky and creates partition data, and an edge extraction process that extracts edge information from the geographic image data A partition extraction processing unit that extracts, as a partition, an edge that forms a polygon based on edge information obtained by the edge extraction processing unit, and each partition obtained by the partition extraction processing unit is used as a map. A first zone shaping processor that shapes the zone into an easy shape, a zone confidence determination processing unit that evaluates the recognition accuracy of each zone obtained by the zone shaping processor as zone confidence, and the zone confidence And a second section shaping processing unit that shapes the section so as to increase the section certainty factor.

  The partition confidence is based on the boundary line confidence based on the extent to which the side of the partition overlaps the edge image, and the closeness between the average color tone of the target partition (for example, farmland) and the color tone of the target partition. It is preferable to obtain it based on at least one of the color difference certainty to be performed and the texture certainty based on the small edge amount in the image.

  When there is adjacent partition data, compare the confidence of the adjacent partition and the current checked current partition. If the confidence of the adjacent partition is lower, use that partition data as is When the certainty of the section is high, it is preferable to replace the side of the current section with an extension of the side of the adjacent section or a parallel one and calculate the certainty of the replaced section.

  The partition certainty is held in a form corresponding to each entry of the partition data, and the surrounding partition data is corrected to a natural form in accordance with the partition data having high confidence. good. If there are multiple candidates for the replacement section, calculate the certainty factor for each of the plurality of sections and determine whether the calculated certainty factor is higher than the original confidence factor. In this case, the replaced section may be used instead of the original section. In addition, when there are a plurality of candidate replacement sections, it may be replaced with a section with the highest certainty among them. Further, it is preferable to output reliability data as a guide for confirmation / correction.

  The present invention may be a method for executing the following steps, a program for causing a computer to execute the steps, and a computer-readable recording medium on which the program is recorded.

  Specifically, an image obtained by photographing the ground surface of a farmland area from a satellite or an aircraft is used as an input, and a polygon is formed from edge extraction processing for extracting edges of roads, rivers, and roads, and data obtained by the edge extraction processing. A section extraction processing unit that extracts edges as sections, a section shaping processing section that corrects each section obtained by the section extraction processing section to a shape that can be easily used as a map, and recognition of each section obtained by the section shaping processing section It is equipped with a block confidence determination processing unit that self-evaluates the accuracy and corrects the data to look more natural, extracts the land data of the farmed area of the photographed area with high accuracy, and also provides confidence data as a guide for confirmation and correction Output together.

  Further features of the present invention will be described in detail by the following embodiments and the accompanying drawings.

  According to the section data creation system of the present invention, when creating an agricultural district map from a geographic image such as a satellite image or an aerial photograph image, creation is performed based on the certainty factor, so that an image suitable for the purpose is created. It is possible to create data. In addition, since the automatically created output result is output, the amount of work to be visually checked and corrected can be efficiently performed, and the creation cost can be greatly reduced. In addition, it is possible to obtain an advantage that the results of these operations are hardly affected by the skill level of the operator.

  Hereinafter, a land parcel data creation system according to an embodiment of the present invention will be described in detail with reference to the drawings.

  FIGS. 1-9 is a figure which illustrates the land parcel data creation system by this Embodiment, In these figures, the part which attached | subjected the same code | symbol represents the same thing, A basic structure and operation | movement are It shall be the same.

  In the embodiment related to the land parcel data creation system described below, for example, an image obtained by photographing a land surface of a farmland area from a satellite is input, and through each image analysis process, the parcel data and certainty data that can be used for the correction are obtained. It is intended to output this. These data are used when creating a farmland map necessary for farmland management.

<System configuration>
FIG. 1 is a functional block diagram showing a schematic configuration example of a land parcel data creation system according to the present embodiment. As shown in FIG. 1, the land parcel data creation system according to this embodiment includes an input / output device 100, a processing device 200 including a personal computer, a workstation, and the like, and a RAM (Random Access Memory) used as a main memory. And a storage device 300 including an auxiliary storage device such as a magnetic disk storage device.

  The input / output device 100 includes an input device 101 including a pointing device such as a keyboard and a mouse, a display device 102 such as a CRT display device, and an output device 103 such as a printer. The input device 101 is used for starting a command or inputting a parameter by a user. The display device 102 and an output device 103 such as a printer are used for presenting images and section data handled by the land section data creation system to the user.

  The processing device 200 includes a section / confidence extraction program 201. The partition / confidence extraction program 201 includes an edge extraction processing unit 202, a partition extraction processing unit 203, a partition shaping processing unit 204, and a partition certainty determination processing unit 205 as program modules. Details of the processing of the partition extraction program will be described later with reference to FIG.

  The storage device 300 stores image data 301, edge data 302, segment data 303, and segment certainty factor data 304. Of these, the image data 301 is an image taken by a satellite or the like, and is data input from outside the system. The other data 302 to 304 are data created during the image analysis process, and the creation of the division data 303 and the division certainty factor data 304 is a feature of the land division data creation system according to the present embodiment. . A detailed example of each data will be described with reference to FIG.

<Data handled>
2A to 2C are diagrams illustrating each data stored in the storage device 300 while using the land parcel data creation system according to the present embodiment. FIG. 2A (a) is an example of image data obtained by photographing farmland from a satellite, and corresponds to the image data 301. FIG. This data needs to be acquired in advance and stored in the storage device 300 before executing the image analysis processing.

  FIGS. 2A (b) and 2A (c) are examples of edge images (images obtained by extracting image contours) created based on the image data (301), and correspond to the edge data 302. FIG. In Fig. 2A (b), only the strong edges of roads (and rivers) are mainly extracted. In Fig. 2A (c), in addition to these, the edges of the tunnels are also extracted. It can be seen that the noise is increased by the amount. As described above, the edge image has different extraction results depending on the edge extraction parameters. The edge data includes a plurality of edge images, and a section extraction process 203 uses these to extract a section.

  FIG. 2B (d) is an example of data indicating a land section and corresponds to the section data 303. One section is treated as one entry, and its contents are expressed by an array of ID, number of vertices, and coordinates of each vertex.

  FIG. 2B (e) is a diagram illustrating an example in which the section data in FIG. 2B (d) is expressed as an image. As shown in the enlarged partial view on the right side (enlarged with the area of Id = 1 as the center), each section specified by Id = 1, 2, 3 connects many vertices T1, T2, etc. It is represented by a polygon TX (see the right figure), and the ID assigned to it, the number of vertices, and the arrangement of coordinates correspond to one entry of the data shown in the table of FIG. 2B (d).

  FIG. 2C (f) is data indicating the certainty level of the segment data (a guideline for self-diagnosis of the correct answer rate), and corresponds to the segment certainty factor data 304. Here, the corresponding partition ID is associated with the partition data 303. In the illustrated system, the partition certainty factor is calculated as, for example, the sum of each element of the boundary line certainty factor, the color difference certainty factor, and the texture certainty factor. Whether the confidence level is high or low is determined based on the obtained value of the partition confidence level.

  The block confidence is based on the boundary line confidence based on the extent to which the side of the block overlaps the edge image, and the closeness between the average color tone of the target block (for example, farmland) and the color tone of the target block It is determined based on at least one of the color difference certainty factor and the texture certainty factor based on the small edge amount in the image.

  FIG. 2C (g) is a diagram illustrating an example in which the partition certainty factor data is expressed as an image, and corresponds to FIG. 2A (c). For example, the difference in the determination of each section is expressed by changing the form of the boundary line that defines the section, and is used as a standard for confirmation / correction. Here, only the boundary line indicating the section specified by Id = 3 is indicated by a solid line, and it is clearly displayed that the reliability is lower than the area surrounded by the boundary line indicated by other broken lines. Details of the relationship between these data and the section / confidence extraction program will be described with reference to FIG.

<Outline of geographic image processing>
FIG. 3 is a diagram showing an outline of the flow of image analysis performed by the land parcel data creation system according to the present embodiment. The land parcel data creation system according to the present embodiment first receives image data 301 prepared in advance, and performs edge extraction processing 202 on the input. Edge data 302 is obtained as an output of the edge extraction process.

  Next, the section extraction process 203 is performed using the edge data 302 as an input to obtain section data 303a. The section data 303a is data obtained by applying the intersection connection method shown in FIG. 4 to the edge data 302, and represents one section as a polygon obtained by connecting a plurality of vectors. The method shown in FIG. 4 will be described later.

  Next, the section shaping process is performed using the section data 303a as an input, and the section data 303a is updated (303b). Further, a part 304a of the section certainty factor data is obtained. By this process, the data faithful to the original image obtained in the section extraction process 202 is updated to data that can be used more easily. More specifically, the section is updated so that the section is represented by a simpler polygon (mostly a quadrangle). During this process, a part of the data that is an element of the partition certainty factor is calculated. The contents of the section shaping process are shown in FIGS.

  Finally, the block certainty factor determination process 205 is performed with the block data 303b and the block certainty factor data 304a as inputs, and the block data 303c and the block certainty factor data 304d are updated. In the process of reference numeral 205, in addition to the data in which mainly the graphic element of the certainty factor is input, here, color and texture information are analyzed and detailed. In addition, the partial shape is corrected in consideration of the arrangement with surrounding sections, and more natural data is created as an agricultural land map. The contents of the partition certainty degree determination process will be described with reference to FIGS. 7, 8, and 9. FIG.

  The plot data and plot confidence data are the target data of this system. In addition to the simple agricultural district drawing data that is easy to use, the self-judgment value that is used as a guide when checking them is displayed as confidence. Has been.

<Details of geographic image processing>
Hereinafter, each of the edge extraction process 202, the section extraction process 203, the section shaping process 204, and the section certainty degree determination process 205 in the geographic analysis system according to the present embodiment will be described in detail.

(1) Edge extraction processing 202
In the edge extraction process 202, a part where the color change is locally large is extracted from the original image as an edge to create edge data.

  Various well-known techniques can be applied as the process of extracting the edge. For example, by applying various edge filters such as the Canny method (see Non-Patent Document 1) and a Sobel filter (see Non-Patent Document 2) to geographic image data, edges such as tunnels can be extracted. By these processes, a portion having a large density change in the image is extracted as an edge. In an image of farmland, it is possible to recognize edges of farmland, roads, roads, and river edges as edges. The extent to which an edge is regarded as an edge depends on the processing parameter, and the edge to be extracted can be limited by adjusting the parameter. In the above Patent Document 2, a method of increasing the extraction accuracy by using a plurality of edge data is proposed. The edge data obtained here is mainly used for processing 203.

(2) Section extraction processing 203
In the partition extraction process 203, a portion that forms a closed polygon by connecting edges from an edge image is extracted as a partition to generate partition data. Here, the straight intersection method (see Patent Document 1) can be used as a method of extracting the sections. FIG. 4 is a diagram showing the flow of the segment extraction process by changing the image. In this section extraction process, a section including the seed point P1 is extracted from the given edge image and the coordinates of one point called the seed point P1 taken in the edge. The seed point P1 refers to the coordinates that serve as the starting point when extracting a section. The seed points P1 can be determined by a method such as interspersed in a grid pattern, randomly arranged, or input in advance by the user.

  As shown in FIG. 4A, when an edge image and a seed point are given, first, straight lines L1 to L3 (n) are extended from the seed point P1 at equal angular intervals in all directions (see FIG. b)). Then, an intersection point Pn of the edge pixels in each straight line and the edge image is obtained, and the obtained intersection point Pn is connected clockwise (or counterclockwise) (FIG. 4C).

  The polygon AR1 extracted by this processing is regarded as one section, and the number of vertices and their coordinates are assigned IDs and registered in the section data (FIG. 4 (d)). The number of vertices Pn of the section depends on the number of straight lines Ln extending from the seed point P1. For example, in FIG. 4, the number of straight lines is four. By increasing the number of straight lines, it is possible to extract a partitioned area that is more faithful to the edge data. Furthermore, by using the method described in Patent Document 2, it is possible to accurately recognize division of a section by a saddle road using a plurality of edge data, and to further improve the extraction accuracy. The section data obtained here is used for the processing after 204.

(3) Section shaping processing 204
In the section shaping process 204, each piece of section data obtained by the section extraction process 203 is shaped into a simple shape such as a square so that it can be easily used as a map. It is preferred that the data used as the farmland map is a simple square, rectangle, or trapezoid rather than a polygon that is faithful to the image, and that the sides of the parcels are parallel or vertical. Shape the data in those shapes to the nearest shape.

5 and 6 are images showing the flow of the section shaping process according to the present embodiment and the contents thereof. In this process, first, the section data obtained by the section extraction process is read as an input (501), and one entry that has not been processed 503 to 50X is selected from the section data stored therein (502). An array of n vertices (coordinates) included in the entry is D _{0 to} D _n . Each line connected in order is regarded as a vector, and the following processing is performed. FIG. 6A illustrates partition data in which vertices are connected in the order of arrangement. The coordinates are labeled D _{0 to} D _n counterclockwise.

  First, the one with the longest distance is selected from two combinations of all vertices. The vertices are defined as Da and Dc (503). Next, Db and Dd having the maximum distance between Da-Db-Dc and Dc-Dd-Da are selected from the vertices of Da to Dc and Dc to Da (504). FIG. 6B shows the result of performing this process on the partition data of FIG. 6A and selecting Da, Db, Dc, and Dd. Here, if the vertices of Da, Db, Dc, and Dd are too close, or if the adjacent sides are almost in a straight line, it can be determined that the figure is a shape that is difficult to approximate to a quadrangle, and the subsequent shaping process is performed. No (505).

  Next, a straight line approximating a series of vectors between the vertices selected in (503) and (504) is obtained (506). For example, in the case of approximating a vector between Da and Db, a method of obtaining an approximate straight line is a method of directly connecting Da and Db, or a least square method so that the sum of distances to the vertices of Da + 1 to Db−1 is closest. There are those for obtaining a straight line and those for correcting the opposite sides to be parallel. In the figure, straight lines obtained by three different methods of solid line L11, dotted line L12, and broken line L13 are shown as corrected lines. A straight line to be actually used can be selected from these. It should be noted that one or more types of methods for obtaining the approximate straight line are sufficient. If there are a plurality of approximate straight lines, the most suitable one is selected from among them (for example, from L1 to L3) in the process of (507). FIG. 6C shows a state in which this process is performed and a plurality of approximate lines are selected for one side of the section.

Next, the certainty factor is obtained for the straight line obtained in (506). If there are several types of obtained straight lines, the degree of certainty for each is obtained and the one with the highest degree is selected (507). For example, the certainty factor is calculated using a parameter obtained by adding A and B with weights as follows.
Certainty for straight line = Wa x A + Wb x B
Here, A is the degree of overlap between the straight line and the edge data, and B is the distance from each vertex. Wa and Wb are weights for each.

  Next, the intersections of the straight lines selected in (507) are obtained and set as Da ', Db', Dc ', Dd', respectively. The original section is replaced with the section having these four points as vertices, and the section data is updated (508). FIG. 6D illustrates the result of performing the processing so far and shaping the sections.

  The processes from 502 to 508 are repeated until all entries included in the partition data are checked (509), and the process is terminated.

  By this processing, each section is modified from a complicated shape having many vertices to a simple section having four vertices that can be easily used for the farmland map. The obtained section data is further corrected in the section certainty factor determination process of (205), and is output together with the certainty factor.

(4) Section certainty factor determination processing 205
In the section certainty determination processing 205, data with higher accuracy and close to manually corrected data can be created based on the section data 303b and the section confidence data 304a obtained by the section shaping process.

  In addition, when the created partition data 303c is used for map creation, certainty factor data serving as an indication of the degree of necessity of correction for each partition is created in order to efficiently perform checking and correction.

  FIG. 7 is a flowchart showing the flow of the section certainty degree determination process, and FIGS. 8 and 9 are diagrams for illustrating and supplementing part of the contents. Hereinafter, details of the processing will be described with reference to FIG. Note that the following processing may be independently obtained for the certainty from a to c.

In this process, the section data and the section certainty factor data are first read as input (701), and then the certainty factor of the section is obtained for each entry of the section data and input to the section certainty factor data (702). For example, the certainty factor is calculated using the following parameters (a) boundary line certainty factor, (b) color difference certainty factor, and (c) texture certainty factor (see FIG. 2) with weights added.
Confidence for partition = Wc × C + Wd × D + We × E
here,
C: Total confidence of each side obtained in process (203)
D: Difference between the average color data in the plot and the standard deviation of the ideal farm area
E: Total amount of edges in a partition Wc, Wd, We: Weight for each Here, the boundary line certainty factor C, the color difference certainty factor D, and the texture certainty factor E in the data illustrated in FIG. is there.

  The “difference between the average of the color data in the section and the standard deviation of the ideal farm area” will be described.

  Information about the ideal farm district image is given in advance by a method such as being stored in the storage device 300 shown in FIG. Therefore, for example, the average and variance (standard deviation) calculated for the color information of all the pixels inside the target section (or a plurality of pixels selected so as to represent the inside of the section) are given in advance. Compare the color information.

  For example, for color information, as the simplest, for each piece of RGB information, R average value / R variance, G average value / G variance, and B average value / B variance inside the section calculate. As a result, in the ideal farmland (or average farmland), the smaller the difference between the R average value / R variance, the G average value / G variance, and the B average value / B variance, Confidence increases. Of course, it is possible to increase the accuracy by a method such as performing a process of adding a value of each color after a high weight is applied to the important difference between the RGB values. For example, when the target is farmland, the average value of G is important, so G is highly weighted. If the target section is a parking lot, the RGB weights are preferably substantially the same. Even in farmland, in winter, the weighting value may be changed according to the time zone, season, etc., by a method such as lowering the weighting of G.

  In addition to RGB, there is a method of obtaining information by HSL (luminance / hue / saturation) as a method of obtaining pixel information. Therefore, this method of obtaining information may be used. The certainty factor may be calculated in consideration of how to obtain a plurality of pieces of information. In the case of information by HSL, the difference in color can be determined by examining the difference between the hue and the saturation in the same way.

  Elements of the certainty factor calculated by this processing are illustrated in FIG. 8 and will be described later.

  Next, one entry that has not been processed (704) to (707) in FIG. 7 is selected from the sections registered in the section data (703), and there is other section data adjacent to the section. Find out if. If there is adjacent section data, the confidence level of the adjacent section and that section is compared (704). If the confidence level of the adjacent section is lower (No), the section data is registered as it is (707), and the process proceeds to checking other section data. When the confidence level of the adjacent section is high (Yes), the side of the currently checked section is replaced with an extension of the side of the adjacent section or a parallel one, and the confidence level of the replaced section is set to (702 It is calculated by the same processing as the processing of). If there are a plurality of candidate replacement sections, the certainty factor of each of the plurality of sections is calculated (705). It is determined whether or not the certainty factor obtained in (705) is higher than the original certainty factor. If the certainty factor is higher than the original (Yes), the subdivision that was replaced is the original one. Is registered instead of (the value in FIG. 2F is also replaced). At this time, if there are a plurality of candidate replacement sections, the confidence of the highest section is replaced (706). On the other hand, if the certainty factor is lower than that of the original section, the certainty factor of the original section is registered without correction (707). An example in which the replacement is performed in this process will be described later with reference to FIG. With the above processing, each section is corrected only in a direction in which the certainty level is higher in accordance with the surrounding high section degree of certainty. By setting the target value of the certainty level so as to be close to the result of manual recognition, corrections that have been manually performed can be automatically corrected based on the parameter of certainty level.

  The processing of (703) to (707) is repeated until the check is completed for all entries included in the partition data (708). If there is still a partition to be checked (No), the process returns to 703. If all checks are completed (Yes), the process is terminated (707).

  FIG. 8 is a diagram exemplifying the level of each element of the certainty factor of the section calculated in (702) of FIG.

  FIGS. 8A and 8B are diagrams showing the level of boundary line certainty. FIG. 8A is a diagram in which the extracted section is superimposed on the original image, and FIG. 8B is a diagram in which the extracted section is superimposed on the edge image. A segment whose side overlaps the edge image has a high boundary line certainty. That is, it can be seen that the right section of two adjacent sections has a higher boundary line certainty than the left section.

  FIG. 8C shows the level of color difference certainty. When the color inside the section is close to the target section, that is, green, which is the color tone of a general farmland, the color difference certainty is high. In FIG. 8 (c), the lower narrow section is close to the color of the farmland, so it can be seen that the section has high color difference certainty. Note that the upper section is a parking lot, which is generally gray and greatly different from farmland, so it can be determined that the section has a low color difference certainty factor.

  FIG. 8D is a diagram showing the level of texture confidence. Many farms have a flat internal color change. Therefore, it is considered that an image with few edges inside the section has high certainty. That is, in the figure, the forest on the left side can be determined as a section with many edge portions in the image and low texture confidence, and the farmland on the right side can be determined as a section with high texture confidence because there are few edge portions in the image. .

FIG. 9 is a diagram illustrating an example of the image / data in which the section is corrected in the process of (706) in FIG. Assume that the data obtained by the section shaping process is shown in FIG. As shown in the figure, the data of the section ID 2 has lower certainty than the data of the other adjacent section IDs 1 and 3. Particularly, the boundary line certainty factor of the side 4 indicated by the reference symbol V1 is very low as “2”, and as shown in FIG. 9B, the side corresponding to the side 4 is redrawn so as to match with another adjacent section. Thus, as indicated by V2, the boundary line certainty factor of the side 4 can be increased to "11". In this way, the certainty level is compared with that obtained by replacing each side of the section, and finally, the data is corrected as shown in the data of FIG. 9B. At that time, for the replaced side, for example, by adding a color, it is easy to indicate which side is replaced.
Through these processes, the section data can be corrected to more natural data as an agricultural land map in accordance with the surrounding section data having a higher certainty level.

Next, a technique related to the technique according to the present embodiment will be briefly described. By the above processing, the certainty factor obtained for each section can be displayed on the display unit in the final step of FIG. 7, for example. By this display, for example, it may be presented that the boundary line certainty of the side 4 of the section ID = 2 indicated by V1 in FIG. 9 is low, and the automatic section data creation processing may be terminated. In this way, for example, the user refers to FIG. 9A and FIG. 9B, and the side 4 of the section ID = 2 has low confidence, and this side 4 is manually set in FIG. It can be seen that correction is required as shown in c). After the correction, it is possible to determine whether or not the correction is successful by automatically obtaining again the certainty factor.
In this way, it is possible to provide an indication of the degree of necessity for correcting the automatically recognized section by calculating the certainty factor.

<Summary>
As described above, according to the block data creation technique according to the embodiment of the present invention, the edge extraction processing unit that extracts edges from geographic image data, and the block extraction processing unit that extracts agricultural land blocks from the edge extraction results And a section shaping processing unit that corrects the obtained section to a shape in accordance with the purpose of use, and a section certainty factor that calculates the certainty factor for each section obtained in the previous processing, and corrects according to the certainty factor A determination unit. Thereby, the map data close | similar to the recognition result by a hand can be created by automating the procedure of farmland map preparation significantly.

  In addition, by providing the “confidence” that self-determined the accuracy of the data output as a result of automatic recognition, it is possible to reduce the cost of visually checking and correcting the automatic recognition result and It is possible to achieve the stability of the accuracy of the result that does not depend on the skill level.

  The present invention is not limited to the embodiments described below. A person skilled in the art can make various changes and improvements to the configuration and functions of the invention described in the embodiments without departing from the scope of the present invention. For example, although the example which aimed at farmland was demonstrated, it is not limited to this. Even if it is not a satellite photograph, it can be applied to a simple image creation process.

  The present invention can be used for partition data creation processing.

It is a functional block diagram which shows roughly the example of 1 structure of the division data creation system by one embodiment of this invention. 2A (a) to 2 (c) are diagrams illustrating various data used in the partition data creation system shown in FIG. FIG. 2B (d) is an example of data indicating a land section, and FIG. 2B (e) is a diagram illustrating an example of the section data of FIG. 2B (d) expressed as an image. FIG. 2C (f) is data showing the confidence level of the partition data (a guideline for self-diagnosis of the correct answer rate), and FIG. 2C (g) is a diagram showing an example of the partition confidence level data expressed as an image. It is a figure corresponding to 2A (c). It is a figure which shows the flow of each process part of the division extraction program in the division data creation system shown in FIG. 1, and its input-output data. It is a figure which shows the example of a process of the data by the division extraction process part in the division data creation system shown in FIG. It is a flowchart figure which shows the flow of a process by the division shaping process part of the division data creation system shown in FIG. It is a figure which illustrates the processing result of the data by the division shaping process part shown in FIG. It is a flowchart figure which shows the flow of a process by the division reliability determination process part of the division data creation system shown in FIG. It is a figure which shows the example of determination of the data by the division reliability determination process part shown in FIG. It is a figure which illustrates the process of the data by the division reliability determination process part shown in FIG.

Claims (6)

A parcel data creation system that analyzes geographic image data obtained from a high sky and creates parcel data,
An edge extraction processing unit for extracting edge information from the geographic image data;
A section extraction processing section for extracting an edge forming a polygon as a section based on edge information obtained by the edge extraction processing section;
A first section shaping processing section that shapes each section obtained by the section extraction processing section into a shape that is easy to use as a map;
A section certainty degree determination processing section that evaluates the recognition accuracy of each section obtained by the section shaping processing section as a section certainty degree;
A partition data creation system comprising: a second partition shaping processing unit that shapes a partition so as to increase the partition certainty factor based on the partition confidence factor.   The partition certainty factor is based on the boundary line certainty factor based on the extent to which the side of the partition overlaps the edge image, and the closeness between the average color tone of the target partition (farmland) and the color tone of the target partition 2. The division data creation system according to claim 1, wherein the division data creation system is obtained based on at least one of a color difference certainty factor and a texture certainty factor based on a small edge amount in an image.   When there is adjacent partition data, compare the confidence of the adjacent partition and the current checked current partition. If the confidence of the adjacent partition is lower, use that partition data as is When the certainty of the section is high, replace the side of the current section with an extension of the side of the adjacent section or a parallel one, calculate the confidence of the replaced section, and calculate the section reliability and The division data creation system according to claim 1 or 2, wherein   The partition confidence level is held in a form corresponding to each entry of the partition data, and the surrounding partition data is corrected to a natural form in accordance with the partition data having high confidence level. Item 4. The section data creation system according to any one of items 1 to 3.   5. The partition data creation system according to claim 1, wherein when there are a plurality of replacement candidate candidates, replacement is performed with the highest partition certainty among them.   The section data creation system according to any one of claims 1 to 5, wherein the section confidence level is output as a standard for confirmation and correction.

Images