JP2000222565A - Method and device for extracting geographical picture change area - Google Patents

Abstract

PROBLEM TO BE SOLVED: To normalize aerial photographs, satellite pictures, or the like different in photographing conditions which are photographs at different points of time and to automatically extract a part where the change like being newly built, collapse, or loss of buildings occurred in a pertinent period. SOLUTION: A building candidate area satisfying specific color and luminance level conditions is decomposed into spectrum components and is expressed with rectangles of N kinds of sizes which have sides parallel with streets, and matching scores peculiar to the geographical picture are calculated from these spectrum components. A successive search nonlinear mapping network using these matching scores is operated to display an area, where matching scores have been reduced, as a picture after mapping, and the area changed which time on the geographical picture is extracted.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a plurality of photographing conditions and photographing times that are widely required in the field of geographic image processing relating to city map updating, disaster prevention image processing, environmental engineering, aerial photography, satellite imagery, and the like. This technology relates to image conversion processing technology that normalizes images and extracts change areas.Especially, it corrects images that have undergone complicated distortions during shooting, while destroying, erasing, and building new buildings, and using land use patterns and vegetation The present invention relates to a technique for extracting a change area of a pollution situation.

[0002]

2. Description of the Related Art Conventionally, a task of extracting a change area of a new building or the like from aerial photographs taken at different points in time is performed by an expert who has a special skill of observing an independent image with left and right eyes. Since it relied on manual work of detecting differences on images one by one, the work required to update a detailed map nationwide became enormous, which became a cost factor for high-precision navigation and the like in recent years. Also, in the event of a disaster such as a great earthquake, it is difficult to shoot at an ideal angle, so manual work of extracting a change area from a photograph containing a large distortion was extremely difficult, which was a factor in stagnation of disaster recovery work.

[0003]

SUMMARY OF THE INVENTION The present invention corrects distortion between two geographic images at different photographing times, including different distortions due to a difference in photographing conditions, etc., as well as newly appearing and collapsing buildings. It is intended to automatically extract a portion where a temporal change has occurred locally such as a change in land use form. As a similar technique for correcting non-linear distortion between two images as described above, for a head MRI image or the like, a “non-linear image conversion device and a standard image calculation method” using a grayscale for each pixel as a matching score ( Patent application number: Japanese Patent Application No. 5-252514; inventor Kosugi et al.)
However, for images based on building shapes with rectangular or square elements, such as geographic images, highly periodic urban areas, and road distribution, matching processing that relies solely on pixel values is not possible. The amount of correction that can be restored is extremely small, and it is practically impossible to apply to a real image having distortion.

[0004]

What is important in the non-linear conversion of a geographic image and the extraction of a time-varying area are a method of calculating a matching score capable of quantifying only an essential difference between two images. Considering the case where a human associates two aerial photographs, a global search based on the shape of a large street or a large building is first performed, and then the collation of fine alleys, small houses, trees, etc. is performed. Is performed. In particular, it is thought that the geometrical characteristics of individual streets and buildings give great information in these works. Also,
In many cases, most of buildings can be decomposed into rectangles parallel to the street. Therefore, in the present invention, a building candidate area that satisfies a specific color or luminance level condition is spectrally-resolved into N types of rectangles having sides parallel to the street, and the following equation composed of the spectral components With the N-dimensional vector FV1, image features around a specific pixel are used. _{FV1 = (P 1, P 2} ... P N) (1) _{Here, P 1, P 2 ... P} N is N having sides in the street main direction
When the rectangles (structural elements) of various sizes are used preferentially in descending order, and when the building candidate area in the target area is filled up, the frequency of each rectangle used represents the spectrum distribution in the order of the rectangle size. And In this spectral representation, the frequency of structuring elements with a small area tends to be overemphasized. This point can be improved by using the following equation in which the effect of the area is corrected. ^{_{FV2 = (L 1 2 P 1}} , L 2 2 P 2, ..., L N 2 P N) (2) _where the L i (i = 1 ... N ) is the length of the large side of the rectangle to the i-th It means to multiply by the square of the side, that is, the coefficient of the area ratio. In addition, in order to reduce the amount of calculation, it is also possible to represent the feature of the image by a scalar feature value F defined by Expression (3). ^{_{F = L 1 3 P 1 +}} L 2 3 P 2 + ... L N 3 P N (3) _{where, L i (I = 1 ...} N) denote the length of the large side of the rectangle to the i-th . This corresponds to the fact that the size is weighted since the square of the side corresponds to the area ratio. That is, the above-mentioned problem is a method of extracting a change area from different geographic images, and specifies a common street between the geographic image A and the geographic image B from the geographic image A and the geographic image B. A process of setting a reference street as
Decomposing the geographic image A and the geographic image B into rectangular figures having sides parallel to the reference street, and generating a matching score from the rectangular figures in the geographic image A and the rectangular figures in the geographic image B A generating step, a step of expanding and contracting the geographic image B based on the matching score, and comparing the rectangular figure of the geographic image A with the rectangular figure of the expanded geographic image B, and as a result of the comparison, And extracting a rectangular figure. Geographic image B based on the matching score
The process of expanding and contracting the rectangular image of the geographic image A and the geographic image B
And generating a movement vector of the rectangular figure of the geographic image B, and expanding and contracting the geographic image B based on the movement vector. In the matching score generation step, the rectangular figure is spectrally decomposed into rectangles of N sizes, an N-dimensional vector is generated, and the N-dimensional vector of the geographic image A and the N-dimensional vector of the geographic image B are compared. This is a matching score generation step of generating a matching score based on the result of the comparison. In the matching score generation step, the rectangular figure is spectrally decomposed into rectangles in which N types of sides are squared, an N-dimensional vector is generated, and the N-dimensional vector of the geographic image A and the N-dimensional vector of the geographic image B are compared. This is a matching score generation step of generating a matching score based on the result of the comparison.
In the matching score generation step, the rectangular figure is spectrally decomposed into rectangles in which N types of sides are cubed, an N-dimensional vector is generated, and the N-dimensional vector of the geographic image A and the N-dimensional vector of the geographic image B are compared. And a matching score generating step of generating a matching score based on the result of this comparison. Further, the above object is an apparatus for extracting a change area from different geographic images, a means for extracting a common street between the geographic image A and the geographic image B from a geographic image A and a geographic image B, Means for decomposing the geographic image A into rectangular figures having sides parallel to the common street;
To a rectangular figure having sides parallel to the common street; a matching score generating means for generating a matching score from the rectangular figure in the geographic image A and the rectangular figure in the geographic image B; Means for expanding and contracting the geographic image B based on the above, and means for comparing the rectangular figure of the geographic image A with the rectangular figure of the expanded and contracted geographic image B, and extracting the inconsistent rectangular figure as a result of the comparison. The geographic image change area extraction apparatus is characterized by having the following. The means for expanding and contracting the geographic image B based on the matching score includes:
It has means for comparing the rectangular figure of the geographic image A and the rectangular figure of the geographic image B to generate a moving vector of the rectangular figure of the geographic image B, and means for expanding and contracting the geographic image B based on the moving vector. . The matching score generating means spectrally decomposes the rectangular figure into rectangles of N sizes, generates an N-dimensional vector, compares the N-dimensional vector of the geographic image A with the N-dimensional vector of the geographic image B,
It is a matching score generating means for generating a matching score based on the result of the comparison. The matching score generation means spectrally decomposes the rectangular figure into rectangles in which N types of sides are squared, generates an N-dimensional vector, and compares the N-dimensional vector of the geographic image A with the N-dimensional vector of the geographic image B. , Is a matching score generating means for generating a matching score based on the result of this comparison. The matching score generating means spectrally decomposes the rectangular figure into rectangles in which N types of sides are cubed, generates an N-dimensional vector, and generates the N-dimensional vector of the geographic image A and the N-dimensional vector of the geographic image B.
This is a matching score generation unit that compares a dimensional vector and a matching score based on the result of the comparison.

[0005]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 shows the flow of the entire process of a method for extracting a temporally changing area of a geographic image using the above feature amounts, and FIG. The flow is shown in FIG. Hereinafter, the method of calculating the feature amount and the flow of the entire process will be described with reference to the drawings. As shown in FIG. 1, the images A and B taken at different points in time are decomposed into structuring elements that take the street direction into account, so that the feature amount F _A defined for each minute area in the image is obtained. and calculating the _{F B.} This specific calculation method is shown in FIG. First, the entire input image is binarized according to the pixel values. Next, Erosion and Dilation are sequentially performed in order to remove noise. Here, Erosion means an operation in which a structuring element (for example, a disk shape) makes a round along the inside of the outline of the plane figure, and the trajectory portion is cut off from the original plane figure. Means that the structuring element makes a circuit along the outside of, and the trajectory portion is added to the original plane figure. By performing both operations, a surface graphic having a diameter within twice the size of the structuring element is regarded as noise and removed, but the original shape of the surface graphic having a certain size is maintained. After the above-described noise removal operation, a fixed (for example, square) neighborhood area is set at the top, bottom, left, and right of the pixel on the image of interest. Subsequently, the surface figure composed of the positive value portions in the neighborhood is spectrally decomposed into rectangular structuring elements having sides parallel to the street direction. The result of the spectral decomposition is quantified using the above formula (1), (2), or (3), and is set as an image feature amount around the pixel of interest. In order to determine the orientation of the rectangle of the structuring element, as shown in the lower part of FIG. 2, when the image is subjected to two-dimensional Fourier transform, and the result is displayed in polar coordinates, the street main point is determined by the peak point on the circumference. The direction shall be determined. Through the above processing, the feature amount _{F A} obtained for the image A, the absolute value of the difference between the feature amount _{F B} obtained for images B | _F A _-F B | by (4), a matching score Define. The matching score is calculated for the part using FV1 or FV2 as the feature quantity by the equation (5) defined for both AB images. {FV (A) -FV (B)} (5) where ‖‖ represents the norm of the vector. The matching score calculated as described above is calculated based on the image B obtained by applying a movement vector for moving the minute area b of the image B by △ x and △ y.
Is evaluated between the fixed images A. _A large number of the motion vectors are generated at the same time, and the matching score | F _A
It shall be adopted as the smallest of _| -F B. (Competition of Motion Vector Candidates) The same operation is performed on pixels around the micro area b, and the motion vectors that have survived the competition are averaged each other within a certain neighborhood range. (Cooperative operation) Each small area of the image B moves based on the averaged motion vector. Next, using this destination as a base point, a motion vector candidate △ x ′, △ y ′ for further movement
Are generated again. The above successive operation is repeated several times to several tens of times. In the latter half of the successive operation, the evaluation criterion is changed to the evaluation of the matching score by the conventional method in which only the pixel value is compared. This is because, as shown in FIG. 3, this method (b) gives a unimodal matching score over a wide search range, and exhibits excellent ability in global optimal solution search, whereas the conventional method (b) This is because a) is not suitable for global search, but has sharp sensitivity in securing the accuracy of final matching. By the above-described motion vector competition and cooperative operation, the image B
Is expanded or contracted so as to best fit the image A, and the difference in shooting conditions is compensated. Moreover, after this operation is completed, a matching score _| F A _-F B | By displays a certain threshold θ or more areas, displaying the new building, collapsed, only a region aging of removal or the like occurs can do.

[0006]

When the features of the geographic image are represented by the vectors of the above equations (1) and (2), the ratio of the latter half of the vector is high in a small building or in an area having a complicated cityscape, while the large building is large. In an area having an object or a simple cityscape, the component of the first half (large structuring element) of the vector becomes large, and it is possible to quantitatively express "complexity of the cityscape" when a human recognizes. In equation (3), the F-number decreases as the complexity of the street increases,
The degree of complexity can be unitarily expressed by a scalar quantity.
The matching score between the two images defined using this feature amount has a unimodal characteristic over a wide range of the amount of deviation between the two images. Although the displacement between the two images due to the difference in the imaging conditions generally occurs two-dimensionally, an example in which the relationship between the amount of the displacement and the matching score is one-dimensionally displayed along a cutting line including the coincident point of the two images. 3 is shown. FIG. 7A shows a matching score according to a conventional method using pixel value evaluation, and FIG. 7B shows characteristics of the matching score according to the present invention.
As described above, in the present invention, it is possible to always achieve convergence to a true solution as an index when executing matching of two images having a large difference. When only the conventional method (a) is used, it is difficult to reach a true solution because it is supplemented by a local minimum solution (local minimum). Examples of the present invention will be described below. 4 and 5 are aerial photographs taken of the same area at one-year intervals, and FIG. 5 is artificially distorted assuming difficult photographing conditions. In this embodiment, as a special case of a rectangle, a square structuring element is used, and processing is performed with N = 3 in the equation (3). FIG. 6 shows the result of expressing a place where the matching score is lowered in a black background in the form of a white blank when the mapping for absorbing the distortion is completed. This result indicates that the area of change (represented in the form of white nucleus in FIG. 7) of a building or the like, which was examined by an expert in detail in detail, was almost accurately detected, and the effectiveness of the present invention in the automatic detection of change over time was confirmed. It is shown.

[Brief description of the drawings]

FIG. 1 is a block diagram showing the overall configuration of an image processing apparatus according to the present invention.

FIG. 2 is a block diagram illustrating a configuration of a geometric feature amount calculation unit based on street direction structuring element decomposition.

FIG. 3 is a diagram showing a relationship between a matching index and a pixel movement amount used in the present invention.

FIG. 4 is a diagram showing an input image to be processed.

FIG. 5 is a diagram showing an input image to be processed.

FIG. 6 is a diagram showing a change candidate area of a building detected according to the present invention.

FIG. 7 is a diagram showing a change candidate area of a building detected by a manual operation of an expert.

Continued on the front page (72) Inventor Shigeru Tsunomoto 6-5 Estate, Kamizuhoncho, Kodaira-shi, Tokyo No.5 Building No. 201, No.5 Mizumotocho (72) Inventor Takeshi Ken Doihara 2751-28 Naruse, Machida-shi, Tokyo F-term 5B057 AA20 BA02 CD11 CE02 CE12 CG05 CH08 DA08 DA20 DB02 DC04 DC30 DC32

Claims (10)

[Claims] 1. A method for extracting a change area from different geographic images, comprising: identifying a common street between the geographic image A and the geographic image B from a geographic image A and a geographic image B; And a step of decomposing the geographic image A and the geographic image B into a rectangular figure having sides parallel to the reference street. The rectangular figure in the geographic image A and the rectangular figure in the geographic image B A matching score generating step of generating a matching score from the following: a step of expanding and contracting the geographic image B based on the matching score; a rectangular figure of the geographic image A and the expanded and contracted geographic image B
And extracting the inconsistent rectangular figure as a result of the comparison. 2. A geographic image B based on a matching score.
The step of expanding and contracting comprises: comparing the rectangular figure of the geographic image A and the rectangular figure of the geographic image B to generate a movement vector of the rectangular figure of the geographic image B; and generating the geographic image B based on the movement vector. The method of extracting a geographic image change area according to claim 1, further comprising a step of expanding and contracting. 3. A matching score generating step includes: spectrally decomposing a rectangular figure into rectangles of N kinds of sizes;
A matching score generating step of generating an N-dimensional vector, comparing the N-dimensional vector of the geographic image A with the N-dimensional vector of the geographic image B, and generating a matching score based on a result of the comparison. The method of extracting a geographic image change area according to claim 1 or 2. 4. A matching score generation step includes: spectrally decomposing a rectangular figure into rectangles in which N types of sides are squared; generating an N-dimensional vector; and calculating an N-dimensional vector of the geographic image A and an N-dimensional vector of the geographic image B. 3. The method for extracting a geographic image change area according to claim 1, further comprising a matching score generating step of comparing and generating a matching score based on a result of the comparison. 5. A matching score generating step includes: spectrally decomposing a rectangular figure into rectangles in which N types of sides are cubed to generate N-dimensional vectors; and generating an N-dimensional vector of the geographic image A and an N-dimensional vector of the geographic image B. 3. The method of extracting a geographic image change area according to claim 1, wherein the method is a matching score generating step of generating a matching score based on a result of the comparison. 4. 6. An apparatus for extracting a change area from different geographic images, comprising: means for extracting a common street between the geographic images A and B from the geographic images A and B; A means for decomposing A into rectangular figures having sides parallel to the common street; A means for decomposing geographic image B into rectangular figures having sides parallel to the common street; A matching score generating means for generating a matching score from the rectangular figure in the geographic image B; a means for expanding and contracting the geographic image B based on the matching score; a rectangular figure in the geographic image A and the expanded geographic image B
A means for comparing the rectangular figure with a rectangular figure, and extracting the rectangular figure that does not match as a result of the comparison. 7. A geographic image B based on a matching score
The means for expanding and contracting is a rectangular figure of geographic image A and geographic image B
7. The method according to claim 6, further comprising: means for comparing a rectangular figure of the geographic image B to generate a movement vector of the rectangular figure of the geographic image B; and means for expanding and contracting the geographic image B based on the movement vector. Geographic image change area extraction device. 8. The matching score generating means spectrally decomposes a rectangular figure into rectangles of N sizes,
Matching score generating means for generating an N-dimensional vector, comparing the N-dimensional vector of the geographic image A with the N-dimensional vector of the geographic image B, and generating a matching score based on a result of the comparison. The geographic image change area extraction device according to claim 6 or 7, wherein 9. A matching score generating means for spectrally decomposing a rectangular figure into a rectangle obtained by squaring N types of sides, generating an N-dimensional vector, and generating the N-dimensional vector of the geographic image A and the N-dimensional vector of the geographic image B. 8. The geographic image change area extraction device according to claim 6, wherein the extraction device is a matching score generation unit that generates a matching score based on a result of the comparison. 8. 10. A matching score generating means, which spectrally decomposes a rectangular figure into rectangles in which N types of sides are cubed, generates an N-dimensional vector, and generates the N-dimensional vector of the geographic image A and the N-dimensional vector of the geographic image B. 8. The geographic image change area extracting apparatus according to claim 6, wherein the apparatus is a matching score generating unit that compares a vector with a vector and generates a matching score based on a result of the comparison.