KR0183335B1 - Surface orientation estimating method of texture image using mathematical morphology - Google Patents

Abstract

The present invention relates to a method for estimating the surface direction of a texture image. In the present invention, the center point of a pattern is obtained by continuous erosion calculation by mathematical morphology, the entire area is divided, and the center segment of the divided subregion is By obtaining a vertical straight line, the infinite point of the entire texture image is calculated and rearranged, and the plane direction of the texture image is calculated using the calculated position of the infinite point and the information of the vertical straight line. By using the divided region information, it is possible to estimate the plane direction with less error even in the image with few linear components.

Description

Surface Orientation Estimation of Texture Images Using Mathematical Morphology

1 is a flowchart of a method of estimating a plane direction of a texture image according to the present invention.

2 is a diagram showing component decomposition in relation to each component in an erosion operation.

3 is a diagram showing a process of finding a center point when performing an erosion operation with a RHOMBUS component.

4 is a diagram illustrating a method of calculating an infinite point of a texture image.

5 to 8 are views showing the estimation result of the plane direction of the texture image according to the present invention.

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for estimating the surface direction of a textured image from changes in a texture image caused by projective distortion, and more particularly to a method for estimating a surface direction of a textured image, using mathematical morphology.

A texture image is an image in which the pattern is distributed, and after JJ Gibson raised the problem of interpreting the texture image by monocular vision in three dimensions (Reference 1, the reference material is described at the end of the detailed description of the invention). The problem of restoring the shape from the texture image has become one of the important research tasks of computer vision.

If the texture image is inclined with respect to the camera's line of sight, various texture gradients can be obtained, and the gradient of the texture image surface is obtained from the gradient of the texture image obtained on the assumption that the pattern is uniformly distributed on the plane. Can be. This problem is to obtain three-dimensional depth information from two-dimensional images. In the conventional methods, the density of edges is calculated based on the assumption that the edges in texture projection correspond to the boundary of the pattern (Ref. 2). Using statistical models from gradient information (Reference 3), Using Fourier analysis to analyze roughness and orientation of texture images (Reference 4), and Method of interpreting fringe distortion with autocorrelation function (Reference 5) ), And the method of clustering and analyzing the horizontal and vertical edge components of the pattern (Ref. 6).

However, the method using the edge density has a drawback that the edge obtained by the operator must correspond to the boundary of the pattern, and the edge not to be removed must be removed. . In the case of using the statistical model from the gradient information, it is difficult to model various fringes, and the method of interpreting the Fourier analysis or the autocorrelation function is local, and has the disadvantage of considering prior information such as isotropy. In addition, the method of clustering and analyzing the horizontal and vertical edge components of the pattern has difficulty in analyzing most natural images composed of complex patterns.

An object of the present invention is to improve the shortcomings of the conventional methods as described above, in order to accurately interpret a texture image composed of a complex pattern, in the present invention, in order to obtain the surface direction from the change of the texture image caused by the projection distortion A new method of using morphology is presented.

Projective distortion considered in the present invention is a perspective effect in which the texture is distantly away from the camera, the perspective effect in which the pattern is uniformly compressed, and the texture is anisotropically compressed in the inclined direction when the texture surface and the image surface are not parallel. Effect.

In order to achieve the object as described above, the present invention, in the method for estimating the surface direction of the texture image, the texture image while performing a morphological erosion operation (erosion operation) to the texture image continuously A first step of finding a center point of each pattern element; A second step of dividing the entire image according to the size of the pattern by using the order of obtaining the center points in the first step; A third step of finding a center line segment of each area by using the center points obtained in the area divided by the second step; A fourth step of calculating an infinite origin by considering the size of the component used in the erosion operation at the intersection point between the average slope of the center line segment obtained in the third step and the vertical line segment perpendicular to the perpendicular line segment; A fifth step of selecting and rearranging center segments of each region from the infinite origin obtained in the fourth step; And a sixth step of calculating the plane direction of the rearranged line segments in the fifth step through mathematical analysis of the perspective transformation.

Hereinafter, the present invention will be described in detail with reference to the accompanying drawings.

1 is a flowchart of a method of estimating a plane direction of a texture image according to the present invention.

First, the first step is a step of determining the center point of the chromosome 10 to find the center point of each fringe that makes up the texture image while continuously performing mathematical morphology erosion operations on the texture image. In the second step, since the size of the fringe far from the observer generally appears to be relatively small, the subdivision step 20 divides the entire image according to the size of the fringe by using the order in which the center points are obtained. The third step is to find the center line segment of each region by using the center points obtained in the divided region (30). The fourth step is to calculate the infinite origin by considering the size of the component used in the erosion operation at the intersection of the average slope of the central line segment obtained in the third step and the vertical line segment perpendicular to the perpendicular line segment. The fifth step is to select and rearrange the center line segment of each region from the infinite origin obtained in the fourth step. Finally, the sixth step is to calculate the plane direction through the mathematical analysis of the perspective transformation of the line segments rearranged in the fifth step (60).

The first step, determining the center point position of the fringe, is accomplished by successively performing morphological erosion operations.

The operator by mathematical morphology has the advantage of preserving the original characteristics of the object shape in the image by selecting and applying the appropriate component to the target image and has the function of a nonlinear filter which is robust against noise. Operators in mathematical morphology include the basic operators dilation (erosion) and erosion and the basic operators consecutive operations (opening) and closing (closing).

In most texture images, the pattern elements have a repeating shape and size, and have a certain constant shade. In the present invention, the acquired texture image is thresholded to an appropriate value, and then unnecessary noise is removed using an opening filter. The position of the center point is determined by performing successive erosions alternately with the binarized images of 3 × 3 RHOMBUS and SQUARE. The continuous erosion with the two components described above is equivalent to the result of erosion with a 5 × 5 circle (CIRCLE). 2 is a diagram showing component decomposition and the relationship between each component in an erosion operation. This method achieves the effect of using RHOMBUS and SQUARE, which decomposes CIRCLE, and reduces the area considering the location of the center point, enabling accurate positioning of the center point even in complex images.

Through the erosion operation, the size of the pattern is gradually reduced, and when the pattern is smaller than the component, the position close to the center of the pattern can be obtained. In the method according to the present invention, since the approximate shape of the projective distorted pattern is used as an ellipse, the circle (CIRCLE) is used, and according to the shape and size of the pattern forming the texture image, various components are appropriately selected. Allows more accurate interpretation of the image.

3 is a diagram illustrating a process of finding a center point when performing an erosion operation with a RHOMBUS among components.

In the case of performing an erosion operation with the RHOMBUS, the weight of the reference pixel is increased by one equation according to the number of pixels in the vicinity of four in consideration of the four neighboring pixels of the reference pixel. For example, if two pixels exist near four of the reference pixel, the value 3 of the value 1 and the weight 2 added together becomes the value of the reference pixel. The pixel having the largest weight is left in the 3x3 mask to determine the center point. If the weight is 5, the erosion operation is continued because it is greater than or equal to the RHOMBUS.

When the result of performing the erosion operation is equal to the third (a) degree, if the weight is given considering the neighboring pixels, each pixel has the weight as shown in the third (b) degree, and the largest weight among them If only the pixel having is left, it is the same as that of FIG. 3 (c). It is the same as FIG. 3 (c). As shown in FIG. 3 (c), when there are two or more pixels having the largest weight in the mask, as shown in FIG. 3 (d), only the pixels having the largest weight in the mask are considered again (the weight 3 (e)), the center point is determined (figure 3 (f)).

When an appropriate component is selected according to the shape of the pattern, an area considering the weight is determined according to the selected component. In the case of erosion to a square SQUARE, the pixels in the vicinity of 8 may be considered.

The center point of each pattern can be obtained by the erosion calculation of the first step as described above, and the center points are determined in order of being far from the observer because of the size change of the pattern itself due to the projection distortion. . Therefore, considering the case where the pattern is irregularly arranged and using the order in which the center points are found and the size of the pattern, the whole area can be divided into small areas.

An area division step, which is a second step of the plane direction estimation method according to the present invention, will be described.

The entire texture image may be divided into subregions by using the center points sequentially obtained by the erosion operation, and the relationship between the subregions in the entire image may be described. Since the texture is not uniformly distributed in the actual texture image, in order to consider such a case, the average of the shortest distances between the points obtained in the process of checking the position of the center point is used. When the point obtained is further away from the calculated average distance, the center point is treated as being included in another area. By labeling each area differently, the area of each area is compared, and only the center point located in the largest area is selected as a meaningful point. Even if the pattern is irregularly distributed, most of the pattern is affected by the projection distortion, so it is necessary to consider the above. In the process as described above, the entire texture image is divided into sub-regions, and the divided region is composed of a pattern of a certain size, and the size is the size of the component used in the erosion calculation.

A step of finding the center segment of each region by using the center points obtained in the divided region, which is the third step of the method for estimating the plane direction of the texture image, will be described.

Hough transforms the significant center points obtained in the region divided by the second step, and this line segment becomes the center segment of each region. Considering the area of each region, find the average slope of the center segment, the line segment orthogonal to it, and the intersection of the two segment components within the subregion. This orthogonal line segment is the tilt information of the texture image. This is based on the study that the three-dimensional analysis of the texture image is possible by obtaining the line segment perpendicular to the direction in which the pattern is most uniformly distributed.

The step of calculating the infinite origin, which is the fourth step of the method for estimating the plane direction of the texture image, will be described.

The condition for interpreting the texture image may be calculated by considering the size of the component used in the erosion operation at the intersection point of the center line segment and the perpendicular line segment of the subregion obtained in the third step. In the present invention, when the entire texture image is composed of a certain size of a pattern, an infinite point is calculated from the size and relative distance information. If the pictograms are affected by the projection distortion, the information about the infinite origin can be obtained through the relationship between the component sizes of each region.

4 is a diagram illustrating a method of calculating an infinite origin of a texture image.

When the pattern forming the texture image is constant, the infinite origin can be calculated from the relative distance between the two projected and distorted patterns. Assuming that the target image is a plane, the information of the infinite origin on one side is coincident. In FIG. 4, when the size of the pattern T ₁ is S ₁ , the size of the pattern T ₂ is S _2, and the relative distance between the two patterns is D, the average value F ₂ for the infinite origin which is the reference point of the entire area is It can be obtained as in Equation (1).

The fifth step is to select and rearrange the center line segments of each region from the infinite origin obtained in the fourth step.

The sixth step is a step of calculating the plane direction of the texture image by performing a mathematical analysis of the perspective transformation of the line segments rearranged in the fifth step.

Perspective transformation of parallel segments in three-dimensional space with an angle of inclination with respect to the z-axis, which is the line of sight, intersects at an infinite origin in the image plane. In three-dimensional space, one line segment L can be expressed as a set of points as shown in Equation (2).

At this time Is passing by, Means the direction segment, to be.

If a point on the line segment L is projected on the image plane, the coordinates (u, v) can be expressed by the following equation (3). Where (x0, y0, z0) is the position of the camera lens at the center of the perspective projection, and f is the focal length.

When the points on the line segment are infinitely far from the center of the lens, λ of Equation (2) is increased to infinity, and the position of the infinite origin can be obtained using Equation (4) below. If b ₃ is 0, all line segments in 3D are parallel to the image plane and there is no infinite point. Using the above equation, the direction component of the line segment of the three-dimensional coordinate system can be obtained by the following equation (5).

The orientation component in Equation (5) can be obtained by using the position of the infinite origin calculated through the rearrangement process, and a relational expression of a vector orthogonal to this component can be obtained by the following equation (6).

The tilt obtained in the region segmentation process can be expressed as in Equation (7). Using Equation (6) and Equation (7), three-dimensional information of the texture image can be calculated.

5 to 8 are diagrams showing the results of estimating the surface direction of the texture image by the surface direction estimation method of the texture image according to the present invention.

FIG. 5 is a composite image (figure 5 (a)) perspectively transformed to look like an exterior wall of a tall building, and FIG. 6 is a perspective view of three-dimensional perspective transformation of Brodatz 'D67 plastic beads (figure 6 (a)), FIG. FIG. 7 is a perspective view of Brodatz's D75 coffee beans in three dimensions (FIG. 7 (a)) and FIG. 8 is a tile image of an actual building wall (FIG. 8 (a)) respectively. The results of applying the method according to the invention are shown.

5 (b) to 8 (b) show the positions of the center points found while continuously eroding. The center point found by erosion twice from angle to circle (CIRCLE) is represented by '+', the center point obtained by erosion three times is represented by '?', The fourth is represented by '×', and the fifth is represented by '○'. It was. As shown in (b) of FIG. 1, it is confirmed that the entire area of the texture image can be divided by the size of the components used in the erosion calculation.

5C to 8C show center line segments in the divided regions of the respective texture images.

5D to 8D show the intersection points of the line segments orthogonal to the center segment in the divided region of each texture image.

5 (e) to 8 (e) show the result of correcting the distance between the centerline and the infinite origin calculated for each texture image.

(F) of Figs. 5 to 8 show the linear components converged to the infinite origin along with the center line segment corrected in each texture image, and the spacing between the convergent straight lines is more than twice the size of the component used in the erosion calculation. As a result, the pattern elements do not overlap each other.

FIG. 9 is a diagram illustrating the edges of a directional pattern having orientation in the target images of FIGS. 5, 6, and 7. In order to find the edge of the pattern, the difference image between the eroded image and the original image was used.

Table 1 below shows the error of each method by comparing the results of the cluster transformation using the vertical and horizontal components of the edges of the respective target images of FIGS. 5 to 8 and the results of the method according to the present invention with the converted actual angles. will be.

In the analysis of the results shown in Table 1, in the case of FIGS. 5 and 8 in which the pattern includes the edges in the linear components, the three-dimensional analysis can be performed within a relatively small error range only by using the cluster transformation using the vertical and horizontal components. It can be seen that. However, it can be seen that in the case of FIGS. 6 and 7 in which the edges of the pattern are irregular or do not contain a linear component, the three-dimensional analysis is difficult by the cluster transformation.

In this case, therefore, the three-dimensional analysis can be performed more accurately using the method according to the present invention. The D67 plastic beads of FIG. 6 had the smallest error since they had a similar composition to CIRCLE. The composite image of FIG. 5 and the tile image of the building wall of FIG. 8 show smaller errors than the results of D75 coffee beans of FIG. Analyzing these results, it can be seen that in the method according to the present invention, using a component similar to the pattern forming the target image, more accurate results can be obtained.

As described above, in the method of estimating the surface direction of the texture image according to the present invention, the center point of the pattern is obtained by continuous erosion calculation by mathematical morphology, the whole area is divided, and the center line segment of the divided subregion and By finding the vertical straight line, the infinite origin of the whole texture image is calculated and rearranged, and the plane direction of the texture image is calculated using the calculated position of the infinite origin and the information of the vertical straight line. Since the divided region information is used, the plane direction estimation with less error can be performed even in the image with few linear components.

Claims (6)

A method of estimating the surface direction of a texture image, the method comprising: a first step of finding a center point of each pattern of a texture image while continuously performing mathematical morphology erosion operations on the texture image; A second step of dividing the entire image according to the size of the pattern by using the order of obtaining the center points in the first step; A third step of finding a center line segment of each area by using the center points obtained in the area divided by the second step; A fourth step of calculating an infinite origin by considering the size of the component used in the erosion operation at the intersection point between the average slope of the center line segment obtained in the third step and the vertical line segment perpendicular to the perpendicular line segment; A fifth step of selecting and rearranging center segments of each region from the infinite origin obtained in the fourth step; And a sixth step of calculating the plane direction of the rearranged line segments through a mathematical analysis of the perspective transformation. The method of claim 1, wherein the component is selected according to the shape and size of the pattern forming the texture image during the erosion calculation of the first step. 3. The method of claim 2, wherein the component is a circle (CIRCLE). The method of claim 1, wherein the area division of the second step is performed when the point obtained by using the average of the shortest distances between the points obtained in the process of identifying the position of the center point in the first step is farther than the calculated average distance. And when the center point is separated from what is included in another area, the center point is achieved by treating it as included in another area. The texture image using mathematical morphology according to claim 1, wherein the center line segment of each region in the third step is obtained by Hough transforming the significant center points obtained in the region divided by the second step. Method for estimating plane direction of. The method of claim 1, wherein the fourth step is to calculate the infinite origin from the relative distance between the two projected and distorted patterns and the size thereof when the pattern of the texture image is constant. Planar estimation method.