JPS6267681A - Estimation system for surface projection of qualitative body from one image - Google Patents

Abstract

PURPOSE:To estimate qualitative projections of the surface of a body by classifying respective picture elements of a gradational digital image of the body surface by plural attitudes and utilizing the relation between those attitudes and the direction of parallel illumination light in image photography. CONSTITUTION:An object image from an image file 11 is stored in an image memory 12, classified through an image classifying device 13 by three kinds, i.e. flat land, a sunny slope, and a shady slope, and stored in an image memory 14. This ternary-coded data is divided in terms of area by an area extracting device 15 and stored as an area number image in an image memory 16. The area number image is inputted to a border line classifying and extracting device 18 together with irradiation direction data in image photography, so that border line images classified by a valley and a ridge are extracted. Further, a qualitative attitude is calculated through an area, border line, and connection point description generating device 110 and a qualitative attitude calculating device 112. Further, smooth and natural qualitative projections of the object image are estimated.

Description

[Detailed Description of the Invention] [Field of Application of the Invention] The present invention relates to a method for estimating the shape of a photographed object from an image, and is particularly suitable for estimating the shape of a photographed object from a single image in the field of resource exploration. Concerning a target object surface relief estimation method.

[Background of the invention]

Conventionally, qualitative object surface relief estimation from a single image has been described in Proceedings of the International Institute of Electrical and Electronics Engineers, Earth Science and Remote Exploration, (-; E-21) using terrain relief estimation as a specific example.
Volume, No. 1 (1983) [EEE Trans.

Geoscjence and RemoTe Sen
sing, V o l G E -21, Na 1
(1983), Wang et al.
tial Reasoning in Remotely
, 5enced Data).

In this example, we will first calculate the pixel value (brightness) of one black and white image.
Divide into dark areas, bright areas, and very bright areas by dividing the boundaries between the areas into valleys (concave boundaries) or It is classified into ridges (convex boundaries). Then, pixels on the valley boundary line are assigned a relative height of O, and pixels on the ridge boundary line are assigned a relative height of 100.Finally, the height of these pixels on the boundary line T is used as a boundary condition to set the relative height on the boundary line. The relative heights of all pixels in the image are estimated by smoothly interpolating the heights of pixels that do not exist.

However, in this example, the region is used only for extracting the boundary line and classifying the boundary line into valleys and ridges, and qualitative slope information of the region itself is not used. Therefore, the valley
It is not possible to assign a uniform height to each ridge, for example O and 100) L, and the accuracy of estimation of ups and downs is low, although it is qualitative.

Furthermore, since a fixed height (0 or 100) is assigned to the constituent pixels of each boundary line, the height may suddenly change step-like at the connection point between the boundary lines, resulting in extremely unnatural estimation of the undulations of the topography. This will be the result.

Furthermore, when dividing an image into regions, there is a problem that the accuracy of division decreases when the size of the image (number of samples) is small or the dynamic range (number of levels of pixel values) of the image is small.

[Purpose of the invention]

An object of the present invention is to provide a method for estimating qualitative object surface relief from a single image, which improves the drawbacks of the prior art and calculates a more natural and likely qualitative object surface relief from a single image. It is in.

[Summary of the invention]

In order to achieve the above object, the present invention first classifies each pixel of one image into several types of attributes, further attributes are added to the regions extracted from the classification results and the boundaries between the regions, and these The qualitative undulations of the object surface are determined by estimating the inclination of the area and the unevenness of the object surface on the boundary line using the relationship between the attributes and the direction of the parallel illumination light at the time of image capture.

Furthermore, when classifying each pixel of an image into several types of attributes, the image is interpolated, image data with a number of samples greater than the number of samples in the original image is generated, and the error of the interpolated value is taken into account. A highly accurate histogram of an image is used, which is calculated by counting the pixel value sections with a weight corresponding to the degree of overlap between the range and each pixel value section of the histogram.

In addition, when determining the qualitative undulations of an object's surface from the unevenness of the object's surface on a boundary line, first, the height is assigned to the boundary line itself and the connection point of the boundary line based on the attributes and connection relationships of the area/boundary line. Assign heights so that the height changes smoothly from these heights to each pixel on the border and smoothly connects to other border lines at connection points.

[Embodiments of the invention]

Embodiments of the present invention will be described below with reference to the drawings.

First, the present invention will be explained in detail. For the image 21 of the target area shown in FIG. 2 obtained by photographing, each pixel of the image is classified into several types of attributes (labels). For example, as shown in Fig. 3, the attributes include a flat land 31 and a sunny slope 32 (facing the lighting direction 34).
, and (the opposite) shady slope 33 are set. For color images, the classification method is the maximum likelihood method (References, Complete Japanese Remote Sensing Research, "Image Processing and Analysis", Imori Publishing Co., Ltd., p. 208-221, 1932).
There are publicly known examples such as May 2006). When the target image 21 is a black and white image, threshold division based on gray level [References, Gonzalez and Wintz
inz) + rDigimol image processing (Digita
l Image Processing) J + Addison-1i1esle
There are known techniques such as y), +977, p, 322-331). When dividing thresholds by density level, set the threshold yourself! l! The histogram of the target image 21 is used for lI selection. In order to calculate a highly accurate histogram, first an interpolation process is performed on the target side a21 to generate an image with a number of samples greater than or equal to the number of samples of the target image 21, and a histogram is calculated from the generated interpolated image. FIG. 4 shows an example of interpolation processing. Figure (a)
shows the spatial arrangement of each sample (pixel) of the input target image 21, and (b) shows the number of samples four times that of the input target image generated by interpolating points with half the sample interval of the input target image shows the spatial arrangement of each sample in the interpolated image. White circles 41 indicate original samples, and black circles 42 indicate samples generated by interpolation. For image interpolation, cubic
There are known techniques such as the convolution method (Reference, for the Japan Remote Sensing Study Group, "Image Processing and Analysis", Imori Publishing Co., Ltd., p. 183-) 84, 1939.
6 years). A histogram is generally expressed as a frequency h (Q) in Q, where Q is the quantum level of image density. When the image density is quantized to 8 bits, α is Ol..., 2
You can get an integer of 55 + I a. Therefore, h (Q) can be calculated by counting the number of pixels with a quantum level of Q among all pixels of the image. However, input target image 2
] Sample values generated by interpolation other than samples are generally real values, and if this real value is converted into an integer, the corresponding level Q is found, and h(Q) is counted, an error will occur in the histogram. Regarding each sample of the input target image, those whose quantum level is Q are Q-0,
It has a signal value between 5 and Q+0.5, with an uncertainty of ±0.5 due to quantization. Therefore, the pixel values generated by interpolation using each pixel of the human-powered target image also include an error of ±0.5. Therefore, when the pixel value generated by interpolation is, for example, 1.8, it is considered to mean any value between 1.3 and 2.3, taking into account errors. Even if it falls on the quantum level Q of the histogram, a is considered to mean the interval from u −0.5 to Q + 0.5. For example, the range of 1.3 to 2.3 is R
= 0°5 to 1.5 interval of 1 and Q, = 1 of 2,
5 = 2, since there is an overlap of ho, 2, and 0.8 with the interval of 5, h(1) has 0.2. Add 0.8 to h (2). As described above, a highly accurate histogram is calculated by counting the pixels generated by interpolation taking into account the error.Next, from the n-value image (n is the type of area attribute) in Figure 3,
Extract a region consisting of pixels with the same attribute value. FIG. 5 shows regions 51 to 56 extracted as a result.

If the images are classified into flat ground 31, sunny slope 32, and shaded slope 33 as shown in Fig. 3, (1) The sunny slope is the surface facing the direction of illumination light.

(2) A shaded slope is a surface facing away from the illumination direction.

(3) As the name suggests, flat land is a flat surface.

By using the relationship between the attribute value of the area, the irradiation direction of the illumination light, and the inclination of the area, if the irradiation direction 34 of the illumination light is set as shown in FIG. The cross-sectional view can be qualitatively estimated as shown in FIG.

As can be seen from FIG. 6, the two regions are in contact with each other in a convex or concave manner. Therefore, one of two types of attribute values, ridge (convex) and valley (concave), is given to the boundary line between two areas.

Figure 7 shows the result of extracting the boundary line between two areas,
It is classified into ridges shown by thick lines and valleys shown by thin lines. Boundary line extraction and classification are performed, for example, in the following procedure.

8 and 9 are enlarged views of the upper right portions of FIGS. 3 and 5, respectively, and the squares correspond to pixels. In the boundary line extraction process, the image of the classification result shown in FIG. 3 and the image of the area number shown in FIG. 5 are sequentially scanned between four adjacent pixels from the upper left. For example, at point 81 in FIG.
Looking at the pixels, from the top left, the areas are 52, 53, 53.53
belongs to. Therefore, point 81 is between area 52 and area 53.
is determined to be a component of the boundary a65 of n■.

Furthermore, from Figure 8, the four adjacent pixels are shaded slope 33 from the upper left.
, sunny slope 32, 32.32, and from the arrangement of the attribute values of these four pixels and the direction of illumination light 34, it is determined that the point 81 is part of a convex boundary line (ridge).

By repeating this process, two
Extract boundaries 71 to 78 between regions, and further extract ridges 79
Or classified into valley 710. Note that areas 55 and 5 in FIG.
The boundary line between 6 and 6 is closed and is originally one, but it can be redivided into two boundary lines 77 and 78 in FIG. 7 by classification into ridges and valleys shown in FIG.

Next, further descriptions are created for the extracted and classified areas/boundaries and the end points (connection points) of the boundaries.

The area number is the attribute value (any of 31 to 33), adjacent area number, etc., and the boundary line is the attribute value (either ridge or valley), constituent point sequence, direction, length, and adjacent 2
These include the number of the area and the number of the boundary line connected at the end point. Also, find the coordinates of the end point (connection point) of the boundary line and the number of the boundary line connected at this point. By creating these descriptions, information about the areas/boundaries themselves and the relationships between them can be easily searched and used.

Based on the created descriptions of areas, boundaries, and connection points, it is possible to estimate the qualitative elevation of the object's surface.

A flowchart of the estimation processing procedure is shown in FIG. To explain using FIGS. 5 and 7 as examples, first, a flat area 51 is found (step 102), and the heights of border lines 61 to 64 touching this are set to 0 (
Step 103). Next, the area 52 adjacent to the area 51 -
For each of 55 (step 106), if a boundary line adjacent to the area includes one whose height is undetermined, the following processing is performed. If the area is a slope, if some of the contacting boundaries have known ridge heights, all ridge boundaries are set to the same height, and the height of valley boundaries is (ridge height m = Set a fixed value (for example, 100) (step 109
). Similar height setting is performed when there is a valley whose height is known (step 1010). Also, if the area is flat, all border lines that touch should be set to the same height (
Step 1011), the height of the boundary line is set by repeating this process.

Note that when the area is included in the area @, 55 like the area 56 in FIG. 5, the height cannot be assigned to the boundary lines 67° and 68 using the above method. In this case, when the area is flat and the area containing it is a slope, the height of the boundary line touching the flat area i is equal to the height of the boundary line touching the area containing it ((
Maximum value + minimum value)/2 sets. Similarly, if the area inside t is a slope, the heights of the ridges and valleys that touch this slope are the (maximum value - minimum value) x - 10 minimum value of the height of the boundary line that is in contact with the area that includes them. and (maximum −
(minimum value)×−10 We will set it to the minimum value. The above processing completes height assignment to all boundary lines.

Next, heights are assigned to the connection points, and heights are assigned to the constituent points of the boundary line. First, the height of the flat ground is adopted as the height of the connection point that touches the flat area. For other connection points, for example, the height of the connecting boundary line (
Set maximum value + minimum value)/2. The height of the boundary line is corrected based on the heights of these connection points, and the heights are assigned to the boundary line constituent points. For example, if the height of the boundary line is 100 and the height of the connecting points at both ends is 0.100, then when the number of points forming the boundary line is n, the height of each point is n-1n-1
Adopt n-1 100. Also, the height of the boundary line is 100°, and the height of the connection points at both ends is 0. In the case of O, 0°n-1n-1n-1 is similarly adopted as the height of each point. The above is the process of allocating heights to the constituent points of the boundary line. Through this process, the height changes smoothly at each point on the boundary line, and the boundary line smoothly connects to other boundary lines at the connection point.

Next, the height of points in the area other than on the boundary line is estimated. The qualitative height of a point in a region has a height only on the boundary line, and other parts are calculated by repeatedly averaging the height using a height image set to 0 or 5o as an initial value. That is, for all points other than those on the boundary line of the height image, replacing the value of this point with the average value of the upper, lower, left, and right pixel values (heights) is repeated until the height image converges.

By the procedure described so far, qualitative estimation of the undulations of the object surface from one image is completed.

FIG. 1 shows the overall configuration of a qualitative terrain relief estimation system from one image as an embodiment of the present invention.

First, the target image is stored in the image memory 12 from the image file 11.
The images stored in are classified into flat areas,
The image memory 14 is classified into three types: FJ facing slope and shaded slope.
The results are stored in .

The image classification device 13 performs interpolation processing on the target image to generate image data with a number of samples greater than or equal to the number of samples in the target image, and calculates the existence range and each pixel value interval of the histogram by taking into account errors in interpolated pixel values. A histogram is calculated by counting the pixel value sections with a weight corresponding to the degree of overlap. Then, two thresholds for classifying the image are calculated using this histogram, and the original target image is classified into three types of regions. FIG. 11 shows the internal configuration of this image classification device 13. In FIG. 11, the histogram of the interpolated image obtained by the image interpolation processing section 131 is calculated by the histogram calculation processing section 132, and from this histogram, the threshold value for quantization is calculated by the threshold calculation processing section 1333. The image processing section 134 classifies the images in the image memory 14 into three types using this threshold and stores them in the image memory 14.

Next, the classification result image stored in the image memory 14 is input to the area extracting device 15, which extracts the area as a closed area after dividing the area, and the area number image as shown in FIG.
is stored in

Next, the area number image and the sunlight (illumination light) irradiation direction data 17 at the time of image capture are input, and the boundary line classification and extraction device 18
classifies the boundary line as either a valley or a ridge and extracts it. The resulting boundary line image is stored in the image memory 19.

Then, the region/boundary line/connection point description creation device 110 inputs the classification result image, the region number image, and the boundary line image, and creates a description of the region/boundary line/connection point. The resulting description is stored in memory 111. As an example of the expression of the description in the memory, expression as a collection of tables as shown in FIG. 12 can be considered. 121, 122, 123
are tables representing descriptions of areas, boundaries, and connection points, respectively; for example, table 122 shows the boundary line number (76).
and component point number, length, direction, adjacent area number (area 54
and area 55), connection boundary line numbers at end points (one boundary line 73 and 74, the other none), attribute value (ridge), height undetermined).

Next, in order to qualitatively determine the altitude of the photographed area, the descriptions of the areas, boundaries, and connection points described above are read from the memory 111 into the qualitative altitude calculation device 12, where the qualitative altitude is calculated, and the target image is A height image representing the qualitative altitude at each pixel is written to the memory 1.13, and heights of boundary lines and connection points are written to the corresponding columns of the table in the memory 111.

Note that the operation of each device is controlled by a control device 114.

Furthermore, in the display device 115, the console 1]-6
By operating the input target image, classification result image,
Images of various inputs, intermediate results, and final results, such as boundary line images and qualitative height images, are displayed on the monitor television 117. Furthermore, the display device 115 provides a stereoscopic display as shown in FIG. 13(a) from the qualitative height image by giving parameters such as a viewpoint, or a cross-sectional view on an arbitrary line as shown in FIG. 13(b). It also calculates and displays.

Furthermore, the display device 115 can refer not only to various images but also to descriptions of areas, boundaries, and connection points in the memory 111, and by using these descriptions, for example, a valley line as shown in FIG. Calculate and display the image shown. Second
The calculation procedure will be explained using the image in the figure as an example. First, by referring to a table 122 in the memory 111 that describes boundary lines expressed in a table format as shown in FIG. 12, a boundary line whose attribute is valley is found. In the example in Figure 7, the boundary line 7
1, 72, 73, 74.77 correspond to this. Next, the two adjacent regions for each found boundary are stored in the table 122.
This time, by referring to the attribute column in the table 121 that describes the region, and extracting the two adjacent regions that are both slopes, the valley line (meaning not only a concave but also a v-shaped valley) is obtained. ). (In the example of FIG. 7, only the boundary line 77). In order to visualize the results, the constituent points of the boundary lines determined to be valley lines are plotted.

As described above, according to this embodiment, qualitative topography can be estimated from one image, and the results can be used in various formats.

It goes without saying that in addition to the undulations of the topography in this embodiment, according to the present invention, it is also possible to qualitatively estimate the undulations of the object surface from a single image taken of the object surface under parallel illumination light. .

〔Effect of the invention〕

As described above, according to the present invention, it is possible to qualitatively estimate the undulations of the object surface from a single image taken of the object surface. As preprocessing, when classifying each pixel of a target image into several types of attributes, a highly accurate histogram can be calculated from an image generated by interpolating diagonal images, so the target image can be classified with high accuracy.

When determining the qualitative undulations of the object surface from the unevenness of the object surface on the boundary line, various values such as 0,100°67.200 are assigned to the boundary line itself and the connection points based on the attributes and connection relationships of the area and boundary line. By smoothly assigning the height of each pixel on the boundary line from these heights, smoother and more natural qualitative undulations can be estimated.

In addition, as an estimation result, in addition to a height image representing the qualitative height of each pixel of the target image, descriptions of regions, boundaries, and connection points can be obtained. It has the advantage that results can be output and used in various formats, including display in 5lit).

[Brief explanation of drawings]

FIG. 1 is an overall configuration diagram of an embodiment of the present invention, FIG. 2 is a diagram showing an input target image, FIG. 3 is a diagram showing a classification result image, and FIG. 4 is a diagram showing a classification result image.
The figure shows an example of the spatial arrangement of each sample of the target image and its interpolated image, Figure 5 shows the region extraction results, Figure 6 is a cross-sectional view taken along line AA' in Figure 3, and Figure 7 Figures 8 and 9 showing the boundary line classification and extraction results are enlarged views of the right side of Figures 3 and 5, respectively, Figure 10 is a schematic flowchart of the height division light processing for the boundary line, and Figure 11 12 is a diagram showing an example of the description-pull expression, and FIGS. 13 and 14' are diagrams showing examples of displaying the relief estimation results.

Claims (3)

[Claims] 1. In a qualitative object surface relief estimation method from a gray-scale digital image taken of an object surface, each pixel in one image is classified into several types of attributes, and further attributes are added to the regions and boundaries between regions extracted from the classification results. By using the relationship between these attributes and the direction of parallel illumination light during image capture, the inclination of the area and the unevenness of the object surface on the boundary line are estimated, and the qualitative undulations of the object surface are obtained. A method for estimating qualitative object surface relief from a single image. 2. When classifying each pixel in the above image into several types of attributes, the image is interpolated to generate pixel data with a number of samples greater than the number of samples in the original image, and the existence range and histogram are Qualitative object surface relief estimation method from one image as described in item 1, which uses a histogram calculated by counting the signal value interval with a weight corresponding to the degree of overlap with each pixel value square interval. . 3. When determining the qualitative undulations of an object's surface from the unevenness of the object's surface on the boundary line, first assign heights to the boundary line itself and the connection points of the boundary line from the attributes and connection relationships of the area/boundary line, and then The height of each pixel on the border line changes smoothly from the height of the border itself and the connection point, and the height is divided into each pixel on the border line so that it smoothly connects to other border lines at the connection point. The first term is a qualitative object surface relief estimation method from one image.