KR101580594B1 - Method for line extraction of high resolution satellite images using regular grid-based hough transform - Google Patents

Abstract

The present invention relates to a linear extraction method of a high resolution satellite image using a regular grid-based Hough transform, which can extract linearity without missing an edge information of an original image by applying a normal grid-based Hough transform technique capable of efficient computation .
A method of linearly extracting a linear image from an image, the method comprising: an image obtaining step (S110) of obtaining an image of a linear region to be extracted from a satellite; A regular grid generation step (S121) of generating a regular grid formed by the predetermined number of pixels in the horizontal direction and the vertical direction in the image, an edge extraction step (S122) of extracting edge information from the image, A normal grid-based labeling step (S130) of assigning a unique label value to an edge according to a position of a regular grid to which the extracted edge belongs; a normal grid-based labeling step (S130) of performing a Huff transform Step S140 is a linear connection that connects the linearities by comparing the co-linearity with the linearities of the adjacent regular grids, (Step S150).

Description

TECHNICAL FIELD [0001] The present invention relates to a method of linearly extracting high resolution satellite images using a normal grid-based Hough transform,

[0001] The present invention relates to a linear extraction method for high resolution satellite images that extracts the linearity of major urban objects such as buildings and roads using high resolution satellite images, and more particularly, to a Hough transform based on a regular grid, The present invention relates to a linear extraction method of high resolution satellite images using a regular grid-based Hough transform capable of extracting linearity without omission of edge information of an original image.

In recent years, as satellite observations have been launched in various countries and spatial resolution has increased, processing techniques for high resolution satellite images have also been greatly improved.

Since pixel-based image processing techniques are affected by noise or small objects in urban areas, it is difficult to accurately extract target objects.

On the other hand, object-based image processing techniques enable recognition of an object itself rather than a single pixel, and various object-based image processing techniques such as image segmentation and feature extraction have been developed.

The linear extraction technique is one of the representative feature extraction techniques used for various purposes in the remote sensing field. The linear extraction technique is widely used as a method for detecting boundaries between buildings and roads and is also used for conjugate element extraction for registration of images of multi-temporal or heterogeneous sensors.

A typical linear extraction technique is Huff transform, and Huff transform is less affected by the noise of the image itself.

However, the Hough transform linear extraction technique has the greatest computational cost and is difficult to apply to large-scale remote sensing images such as aerial photographs or satellite images. Conventional studies have performed additional processing for repeated search of the area to be transformed or connection to the adjacent line in order to apply the Hough transform to the remote sensing image, which is a disadvantage that the existing high computational cost problem is exacerbated.

Meanwhile, the following prior art documents include a patent on the automatic extraction method of building outline using airline data and a comparative study on the results of GDPA and Hough transformation processing in linear feature extraction using satellite image information have.

The paper has the disadvantage that it can not extract the linearity representing the region by extracting the linearity by applying the variables to the entire image at a time, and does not consider post-connection between the extracted linearity.

KR10-0963651B

 Korean Journal of Remote Sensing, Vol.20, No.4, 2004, pp261 ~ 274

SUMMARY OF THE INVENTION The present invention has been made to solve the above problems, and it is an object of the present invention to provide a method and apparatus for efficiently extracting a linear shape by eliminating a repeated optimization process for the size and position of a transform region, And a method of linearly extracting high resolution satellite images using a regular grid-based Hough transform.

Another object of the present invention is to provide a linear extraction method of a high-resolution satellite image using a normal grid-based Hough transform capable of extracting a line having a high degree of similarity with edge information of an original image and minimizing a missing line in an object in the image .

According to another aspect of the present invention, there is provided a method of extracting a linear shape from a video image, the method including: obtaining an image of a linear region to be extracted from a satellite; A regular grid generating step of generating a regular grid formed by the predetermined number of pixels in the horizontal direction and the vertical direction in the image; an edge extracting step of extracting edge information from the image; A regular grid-based labeling step of assigning a unique label value to an edge according to the position of the normal grid to which the extracted edge belongs; and performing a Hough transformation on an edge having the same label value to extract a linear shape in each regular grid A Hough transformation step of transforming the normal grid into a linear matrix of one of the normal grids, The regular grid, and And a linear linking step of linking linear lines by analyzing whether the adjacent regular lattice linear shapes meet the co-linearity.

The regular grid generation step is characterized by generating a regular grid array with a predetermined number of pixels in the horizontal direction and the vertical direction in the video image and assigning a unique label value according to the position of the regular grid .

In the edge extracting step, an edge is extracted using a Canny operator.

The Huff transform step performs Hough transform on an edge having the same label value, finds a maximum accumulation point in the Huff transform space, and extracts a representative linear shape of the corresponding label grid.

In the linear linking step, at least one linear pair of the normal grid including one of the grid lines extracted in the Hough transform step and corresponding linearities of the corresponding grids surrounding the regular grid is within an angle difference determined by each other, And the linearity of the edge extracted from the adjacent two regular grids is integrated with each other if the shortest distance is within a predetermined ratio to the normal grating size.

The conditions for integrating the linearities extracted from the adjacent two normal grids in the linear connection step are that the linearities extracted from the adjacent two regular grids are within 10 degrees of each other and the shortest distance between the two linear end points is 1 / 3. ≪ / RTI >

According to the method of linearly extracting high-resolution satellite images using the regular grid-based Hough transform according to the present invention having the above-described configuration, it is possible to reduce the computation speed .

Also, it can be used as a basic linear extraction algorithm for providing cues of various application fields such as object extraction and conjugate element extraction using high resolution satellite images because of high similarity with the original image edge.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a flowchart illustrating a method of linearly extracting high resolution satellite images using regular grid-based Hough transform according to the present invention.
FIG. 2 is a conceptual diagram of a linear connection using grid neighbor information in a linear extraction method of high resolution satellite images using regular grid-based Hough transform according to the present invention.
FIG. 3 is an image of a building area photographed by Worldview-2 satellite in order to verify a linear extraction method of a high resolution satellite image using a regular grid-based Hough transform according to the present invention.
FIG. 4A is an image obtained by linearly extracting the image of FIG. 3 by a linear extraction method of a high resolution satellite image using a normal grid-based Hough transform according to the present invention.
FIG. 4B is an image obtained by linearly extracting the image of FIG. 3 by a linear extraction method using standard Hough transform. FIG.
FIG. 5 is an image of a road region taken by a Geoeye satellite in order to verify a linear extraction method of a high resolution satellite image using a regular grid-based Hough transform according to the present invention.
FIG. 6A is an image obtained by linearly extracting the image of FIG. 5 by a linear extraction method of a high resolution satellite image using a normal grid-based Hough transform according to the present invention.
FIG. 6B is an image obtained by linearly extracting the image of FIG. 5 using a linear extraction method using standard Hough transform. FIG.

Hereinafter, a method of linearly extracting a high-resolution satellite image using the normal grid-based Hough transform according to the present invention will be described with reference to the accompanying drawings.

As shown in FIG. 1, a method of linearly extracting a high resolution satellite image using a regular grid-based Hough transform according to the present invention includes an image obtaining step (S110) of obtaining an image of a linear region to be extracted from a satellite, A regular grid generating step (S121) of generating a normal grid formed by the predetermined number of pixels in the horizontal direction and the vertical direction, an edge extracting step (S122) of extracting edge information from the image, A normal grid-based labeling step (S130) of assigning a label value to an edge according to a position of a normal grid to which the target grid point belongs; a Huff transform step (S140) of Hough transforming an edge having the same label value; (S150), which are connected to each other according to predetermined collinear conditions.

In the image acquiring step (S110), a high-resolution image obtained by photographing a target area from a satellite is input. The image is a high-resolution image obtained by photographing a target area in order to extract a linear shape of a ground object.

The regular grid generation step S121 generates a regular grid array with a constant size in the image.

The sum of the sizes of the individual regular grids is made equal to the size of the input image.

In addition, a predetermined number of pixels form one regular grid. For example, the regular grid may be formed with 10 pixels in a horizontal direction and a vertical direction in a size of 10 × 10 pixels. Generally, in a high resolution image, 10 pixels corresponds to a size of 5 to 10 meters, which is usually small enough to detect objects of interest (buildings, roads, etc.) in a high resolution image.

Thus, by using a regular grid, the size of the regular grid can be fluidly converted according to the scale of the object to be detected. That is, the size of the normal grid can be generated differently considering the scale of the image.

On the other hand, when a plurality of regular grids are generated so as to be equal to the size of the input image, each normal grid constituting the image is set to have a unique label value. The label value increases sequentially from left to right, and sequentially increases from the top to the bottom. For example, if 100 × 100 regular grids are generated, the first row is 1, 2, 3, ... , 100, and the second row has labels 101, 102, 103, ... , 200 labels, and in this way the last row is labeled 9901, 9902, 9903, ... , ≪ / RTI > 10000. The method of labeling the regular grid is not limited to the above example, and can be changed to have a unique label for each individual regular grid.

In the edge extracting step S122, edge information used for linear extraction in the image is generated.

To this end, in the edge extraction step S122, a canny operator, which is one of the edge extraction operators, is used.

The Canny operator determines the intensity (m) and the direction (o) of the edge based on the amount of change G of brightness values of the x and y axes as shown in equation (1).

Equation (1):

In the regular grid-based labeling step (S130), different label values are assigned to the edges according to the position of the regular grid to which the extracted cane edge belongs. The label is given to the edge according to which lattice the position of the edge generated in the edge extracting step belongs to, based on the label assigned independently for each regular grid in the regular grid generation step.

The Hough transform step S140 performs a Hough transform on an edge having the same label value among the edges. The Hough transform is one of image processing techniques for detecting a linear component at a binary edge. When the Hough transform is performed, a linear shape is extracted in each regular grid.

In the Hough transform step S140, the Hough transform is performed such that the x and y coordinates of the edge pixels constituting the line are calculated by an angle (?) Formed by the distance (?) At the origin and the normal line direction of the x- Lt; / RTI > The connectable linear set for each edge pixel is a sine curve in the p-θ plane, and the peak of the cumulative sine curve is the representative line of the corresponding label value.

Equation (2):

(Where θ is the angle formed by the line connecting the origin and the line vertically to the x axis in the x, y plane, and ρ is the vertical distance between the origin and the line)

In the linear connection step (S150), the extracted linear shapes of the regular grids are compared with the linear shapes of the adjacent regular grids, and when the collinear conditions are satisfied, they are connected and updated.

That is, in the linear connection step S150, if at least one linear pair satisfies a collinear condition with respect to the linearities of the corresponding available lattices surrounding the arbitrary lattice linear form, the linearities extracted from the adjacent two normal lattices are integrated do.

The adjacent grid can be classified into nine kinds according to the position of the grid, and the criterion for dividing the type of the adjacent grid is the number and direction of the adjacent grid. The adjacent grid can be divided into upper left corner, upper right corner, lower left corner, lower right corner, upper row, lower row, left column, right column, and the remaining cases occupying most of the other, depending on the position of the center regular grid have.

At this time, the number of adjacent lattices adjacent to one of the grids exists from 3 to 8, and connects the linear lattice satisfying the condition of the linear pair corresponding to the reference lattice.

FIG. 2 is a conceptual diagram of a linear connection using the grid adjacent information in the linear connection step (S150). In the linear connection step (S150), if two specific linear matrices satisfy the collinear condition for all the corresponding pairs of gratings adjacent to the center grating, two linear matrices are integrated with each other.

For the collinear condition, if the angular difference between the adjacent lattice line shapes is less than the predetermined angle and the shortest distance among the two end points is less than the predetermined ratio in comparison with the lattice size, the two linear types are integrated. The reference of the angle difference may be 10 degrees, and the ratio of the distance between the both end points to the grid size may be 1/3. At this time, the angle and distance conditions for integrating the two linear shapes can be changed as needed.

The integration of the linearity uses two linear regression lines, the coordinates are updated through the vertical projection to the regression line of the outermost point, and the linear is connected. When a linear is connected, the linearity of the corresponding grid pair is replaced by the update linear. This linear linking operation is performed for all grids.

Since the linear extraction method of the high resolution satellite image using the regular grid-based Hough transform according to the present invention has a simple unidirectional bottom-up structure using the regular grid, Since there is no top-down operation structure and utilizes predefined adjacent grid information, it is not necessary to calculate the distance and direction of all adjacent linearities, and the operation speed increases.

FIG. 3 shows an image of a building area photographed by Worldview-2, which is a satellite for shooting a high-resolution image. According to the method for linearly extracting high resolution satellite images using the normal grid-based Hough transform according to the present invention, an image as shown in FIG. 4A can be obtained, and an image as shown in FIG. 4B can be obtained by performing a standard Hough transform .

In the standard Huff transformation, general labeling is applied to edges generated in the edge extraction step (S122) and Huff transformation is performed. A sufficient number of accumulation points is selected to extract a linear shape and then a redundant linear shape is removed. At this time, the number of accumulation points can be four.

Meanwhile, FIG. 5 shows an image of a road area photographed by Geoeye, which is another satellite for photographing high-resolution images. Similarly, according to the method of linearly extracting a high resolution satellite image using the normal grid-based Hough transform according to the present invention, an image as shown in FIG. 6A can be obtained and an image as shown in FIG. 6B can be obtained by performing standard Hough transform.

The images of FIG. 4A and FIG. 6A using the linear extraction method of the high resolution satellite image using the normal grid-based Hough transform according to the present invention showed a faster calculation speed than the images not having the linear image. That is, when the normal grid-based Hough transform according to the present invention is applied to the images of FIGS. 3 and 5, the images of FIGS. 4A and 6A are obtained, and the processing time is 0.650 seconds and 1.707 seconds, respectively. On the other hand, in the same hardware environment, standard Hough transform is applied to the images of FIG. 3 and FIG. 5, respectively, and the processing time of 1.053 seconds and 2.919 seconds is required to process the images of FIGS. 4B and 6B, respectively. The image processing time may vary depending on the hardware construction environment, but it can be seen that the normal grid-based Hough transform according to the present invention is faster than standard Hough transform in the same hardware environment. The normal grid-based Hough transform scheme of the present invention showed 62% and 71% faster computation speeds than standard Huff transforms in the Worldview-2 building area and Geoeye road area, respectively.

In addition, there is an advantage in that the degree of similarity to the white pixels in FIG. 4 and FIG. 6, which is the edge information of the original image, is high and linearity is extracted without missing in the road of the minor axis and curve of the building. On the other hand, in the case of the standard Hough transform, as shown in the red ellipses in FIGS. 4B and 6B, many shortcomings occur in the short-axis and curved roads of buildings.

S110: Image acquisition step
S121: regular grid generation step
S122: edge extraction step
S130: Regular grid-based labeling step
S140: Hough conversion step
S150: linear connection step

Claims (6)

A method for extracting a linear shape from a video image,
An image obtaining step of obtaining an image of a linear extraction subject area from a satellite,
A normal grid generating step of generating a regular grid formed by the predetermined number of pixels in the horizontal direction and the vertical direction in the image,
An edge extraction step of extracting edge information from the image;
A normal grid-based labeling step of assigning a unique label value to an edge according to a position of a normal grid to which an edge extracted from the image belongs;
A Hough transform step of performing Hough transform on an edge having the same label value to extract a linear shape in each normal lattice,
And a linear linking step of linking the linear shapes by analyzing whether the linearity of any of the normal grids and the linear shapes of the regular grids adjacent to the regular grating satisfy the co-linearity of the regular grids. A method for linear extraction of high - resolution satellite images using regular grid - based Hough transform. The method according to claim 1,
Wherein the normal grid generation step comprises:
A regular grid array is formed by the predetermined number of pixels in the horizontal direction and the vertical direction in the image,
And a unique label value is assigned according to the position of the regular grid generated in the regular grid generation step. The method according to claim 1,
And extracting an edge using the Canny operator in the edge extraction step. 2. The method of claim 1, wherein the edge extraction is performed using a Canny operator. The method according to claim 1,
Wherein the Huff transform step comprises:
Wherein the Huff transform is performed on an edge having the same regular grid label value, and a representative accumulation point of the corresponding label grid is extracted by finding a maximum accumulation point in the Huff transform space. Extraction method. The method according to claim 1,
In the linear connection step,
For a normal grid including any one of the grid lines extracted in the Hough transform step and linearities of corresponding grids surrounding the normal grid,
If at least one linear pair is within a predetermined angle difference and the shortest distance between both end points is within a predetermined ratio with respect to the normal grid size generated in the normal grid generation step,
And the linearities extracted from the two adjacent regular grids generated in the regular grid generation step are integrated with each other. The linear extraction method of high resolution satellite images using the regular grid-based Hough transform. 6. The method of claim 5,
The conditions for integrating the linearities extracted from the adjacent two normal grids in the linear connection step are as follows.
The linear extraction of the high-resolution satellite image using the regular grid-based Hough transform is characterized in that the linearities extracted from the adjacent two regular grids are within 10 degrees of each other and the shortest distance between the two end points is within 1/3 of the normal grid size. Way.

Images