KR20160150455A - Method and apparatus for extraction of edge in image based on multi-color and multi-direction - Google Patents

Abstract

A method for extracting the outline of an image based on multiple colors and multiple directionalities comprises a color defining step of defining multiple colors by a plurality of colors constituting a color image and an infrared image received from a multi-color sensor; a direction defining step of defining multiple directionalities by a plurality of directions including the x, y, and diagonal directions of an image; a derivative calculation step of calculating a derivative of each of predetermined multiple directionalities, for an image of each color of predetermined multiple colors; a slope representative value calculating step of computing a derived derivative and calculating a slope representative value in the image; an outline refinement step of obtaining an outline as an extreme value from the derivative and the slope representative value for an outline area where similar pixels are gathered and removing the outline except the extreme value; and an outline detection step of determining and detecting an outline by using adaptive multiple thresholds computed based on a histogram in the image.

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and apparatus for extracting contours of an image based on multiple colors and multi-

The present invention relates to a method and an apparatus for extracting edge information of an image based on multiple colors and multi-directionality, and more particularly, to a method and apparatus for extracting edge information of an image, (Eg, RGB + IR) and multi-directional slope (frequency) to enhance the extraction accuracy of the contour line and improve the extraction reliability and speed of the contour line using the adaptive threshold based on the information of the histogram of the previous frame And to a method and apparatus for extracting contours of an image including its preprocessing step (removing noises and emphasizing boundaries), and post-processing steps (removing protrusions, perpetuating, and removing noise).

Due to the rapid development of image processing technology, contour extraction technology has become an issue in various fields. This technique can be applied to a variety of applications through template matching using contour information, image-based recognition technology, and stereo matching technology. For example, the following application fields can be cited.

(1) Template matching - Object recognition, fingerprint recognition,

(2) Image-based recognition technology - Face detection, face recognition,

(3) Stereo matching - 3D display

In the prior art of this field,

(1) Sobel filter technology

(2) Prewitt filter technology

(3) Roberts filter technology

(4) Laplacian filter technology

(5) Canny filter technology

And the like.

In these prior arts, the inclination representative value in the image is extracted using a single sensor, a single direction (based on X direction and Y direction), in the process of extracting the contour information. Here, even if a variable environment change occurs, a method of extracting a contour line by applying a fixed threshold value is performed.

Patent Publication 10-2004-0035981

However, according to the prior art, there is a problem that (1) the accuracy of the extracted contour is low, and (2) the result is sensitive to a change in the environment.

In other words, contour information can be extracted even in a single color image based on the conventional X direction and Y direction. However, if the background is complicated and a variable change such as a change in illumination occurs, There was a problem of difficulty. Therefore, the processing speed needed to maintain accuracy (without loss of speed) while maintaining real-time processing.

An object of the present invention is to provide a technique for improving the extraction accuracy of image contour information based on multiple colors and multi-directionality and improving the reliability of extracted contour lines using an adaptive threshold value based on histogram information in the image will be. Hereinafter, the image results of the conventional contour extraction technique are relatively compared and analyzed.

According to another aspect of the present invention, there is provided a method of extracting a contour of an image based on multiple colors and multi-directionality, A color defining step of defining multiple colors by a plurality of colors constituting an infrared image; a direction defining step of defining multiple directions by a plurality of directions including x, y and diagonal directions of the image; for color images, and the derivative calculation step to obtain the derivative for each direction in a given multi-directional, operation processing the obtained derivative, video output stage within the slope represents the slope representative values to obtain a value, and a contour which is similar to the pixel gathered region , The extreme outline is obtained from the derivative and the slope representative value, and the outline other than the extremum is removed W and contour refinement step of a thin contour, and on the basis of the image in the histogram using the adaptive multi-threshold operation, it characterized in that it comprises an edge detection step of determining and detecting a contour.

Here, in the calculating step prior to said derivative, for the color image and an infrared image received from the multi-color sensor, removes noise, and detected after pre-treatment (Pre -processing) step or said detection step the contour emphasizing the boundaries results Eliminates overhangs, perpetuity, and noise It is preferable to further include a post - processing step .

The arithmetic processing for obtaining the inclination representative value may be an Euclidean gamma or a square root processing of the sum of the squares of the respective derivatives or a summation of the absolute values of the respective derivatives.

In the contour refinement step, considering the x and y directions, (1) case 0xx is defined as a negative part, case 1xx is defined as a positive part, (2) case x0x (Xx1) is defined as a negative part, and case xx1 is defined as a positive part, the case xx1 is defined as a positive number part, and (3) the case xx0 is defined as a positive part, It is preferable to select the case in consideration of the derivative of the multi-directionality, extract the difference value from the slope representative value in the image, and calculate the extreme value.

In the contour detection step, the histogram may be generated, the total number of pixels of the contour may be calculated, and a predetermined ratio of the total number of pixels of the contour may be calculated as a threshold value.

The histogram is preferably a histogram of a previous frame of a frame for calculating a threshold value.

In the pre-processing step, it is preferable that the weighted median filter is applied in which the weighted kernel to which the Gaussian distribution is applied is applied as the frequency.

In the post-treatment step, it is preferable to carry out a removal treatment for removing protrusion and noise by performing expansion after erosion, and a filling treatment for continuing the erosion after expansion.

Meanwhile, an apparatus for extracting contours of an image based on multi-colors and multi-directionality according to the present invention is an apparatus for extracting contours of an image based on multi-colors and multi-directionality. The apparatus extracts contour lines of a color image and an infrared A direction defining unit for defining multiple directions by a plurality of directions including an x, y direction and a diagonal direction of the image; and a color defining unit for defining multiple colors by a plurality of colors, for, and to obtain the derivative for each direction of the multi-directional derivative calculation unit determined, by calculation processing of the obtained derivative, image tilt to obtain within the slope representative value expected pyogap calculation unit and, for the contour region with the similar pixels together, the derivative and obtaining the extreme value of the slope contour from the central value, the outline of the thin contour by removing the contour of the non-peak Using the refined portion and, within the image histogram the adaptive multi-threshold operation based on, it characterized in that it comprises a contour detector for determining and detecting a contour.

Here, the derivative calculation unit operation prior to, for a color image and an infrared image received from the multi-color sensor, remove noise and pre-emphasizing a boundary (Pre -processing) unit or, the edge detection operation in a later step, the detection Eliminating overhangs, perturbations, and noise in the results It is preferable to further include a post-processing unit .

The arithmetic processing for obtaining the slope representative value may be Euclidean gamma or the square root processing of the sum of the squares of the respective derivatives, or may be a summation processing of the absolute values of the respective derivatives.

In the contour refinement step, considering the x and y directions, (1) case 0xx is defined as a negative part, case 1xx is defined as a positive part, (2) case x0x (Xx1) is defined as a negative part, and case xx1 is defined as a positive part, the case xx1 is defined as a positive number part, and (3) the case xx0 is defined as a positive part, The case may be selected in consideration of the derivative of the multi-directionality, and the difference between the case and the inclination representative value in the image may be extracted to calculate the extremum.

In the contour detection step, the histogram may be generated, the total number of pixels of the contour may be calculated, and a predetermined ratio of the total number of pixels of the contour may be calculated as a threshold value.

The histogram may be a histogram of a previous frame of a frame for calculating a threshold value.

In the pre-processing step, it is preferable that the weighted median filter is applied in which the weighted kernel to which the Gaussian distribution is applied is applied as the frequency.

In the post-treatment step, it is preferable to carry out a removal treatment for removing protrusion and noise by performing expansion after erosion, and a filling treatment for continuing the erosion after expansion.

Meanwhile, the computer-readable medium of the present invention is a computer readable medium having a plurality of color images and a plurality of infrared images constituting an infrared image, A color defining step of defining multiple colors by color; a direction defining step of defining multiple directions by a plurality of directions including x, y and diagonal directions of the image; the calculation process of the derivative and the derivative calculation step of obtaining a derivative, obtained for each direction of the multi-directional, for picture and calculating steps in the slope represents the slope representative values to obtain a value, the contour region with the similar pixels together, the derivative and the slope representative obtaining a peak value from the outline, by removing the outline of the non-peak contour detail for a thin outline Readable media screen; and imaging using my histogram an adaptive multi-threshold operation on the basis of, the computer storing a program for executing the steps of: characterized in that it comprises an edge detection step of determining and detecting a contour to be.

The contour detection method of the image of the present invention is a method of detecting the contour of an image. In order to divide contours into three types, i.e., strong, wek, and four, applying a threshold value determining step, a high threshold and a low threshold value of the predetermined threshold to determine the features that the recursive scheme a contour including a contour detecting step of determining and the detection.

On the other hand, projecting within the contour extraction in the image of the present invention, constant, noise reduction method, projecting in the edge detection in the image, continuously, as a method for removing noise, removal of performing the expansion after erosion in order to eliminate the projecting and noise processing after expansion to give a phase lead and a constant characterized in that the fill includes a processing step of performing erosion.

Here, the erosion is performed when a contour line exists in the center pixel of the mask and the number of pixels of the outline in the mask is equal to or larger than [the mask width], and the erosion occurs when the mask center pixel in the two- It is preferable to remove it if it is smaller or equal.

The expansion is performed when there is no contour line in the center pixel of the mask and the number of pixels of the outline in the mask is equal to or larger than [the mask width - 1], and when the center pixel of the mask in the two-dimensional histogram becomes the maximum, Is preferably unchanged.

According to the present invention, an image processing technique that improves the extraction accuracy of image contour information based on multiple colors and multi-directionality and improves the reliability of the extracted contour using an adaptive threshold value based on histogram information in the image / RTI >

Figure 1 shows a block diagram for extracting contours based on multiple colors and multiple directions.
Fig. 2 shows a process of a conventional median filter.
[Figure 3] shows the process of the weighted median filter of the present invention.
4 shows the results of comparison with a median filter, a weighted median filter, and a conventional filter.
FIG. 5 shows a model considering multi-directionality and a definition of a derivation operation process for the model.
FIG. 6 shows the definition of the Euclidean norm and the process for processing a negative number.
FIG. 7 shows the results of the derivative taking into account the multi-directionality and the result of the slope representative value (Magnitude).
FIG. 8 defines 8 conditions for edge thinning and a calculation method therefor.
9 shows the processing before and after the contour refinement.
Figure 10 shows definitions for three types of contours (Strong, weak, necessary).
11 shows a process of contour detection.
12 is a timing chart for generating a histogram in the contour detection step.
[Figure 13] shows comparison results with a single threshold value and multiple threshold values in comparison with conventional results.
[Figure 14] defines features (protrusion, noise, continuation) for contour correction.
[Fig. 15] shows conditions and processing procedures for erosion and dilation.
Fig. 16 defines the characteristics of the outline to be extracted.
FIG. 17 shows a maximum value extraction process based on a two-dimensional histogram.
[Fig. 18] shows a processing procedure for contour correction.
Fig. 19 shows the result of contour correction.

Hereinafter, the present invention will be described in detail with reference to the accompanying drawings. However, for elements having the same function by the same structure, the same reference numerals are retained even if the drawings are different, and detailed description thereof may be omitted.

The present invention relates to a method and apparatus for extracting contour information (contour) of an image based on multiple colors and multi-directionality. A block diagram of the block is shown in Fig. However, the technique of the present invention is not limited to a method of extracting a contour, a software such as a computer in which the method is implemented, a recording medium in which the computer software is stored, a hardware device in which the method is block- Can be realized. Hereinafter, the method will be mainly described.

On the basis of multi-color and multi-directionality of the invention, a method of extracting the outline information (outline) of the image, a color definition phase, direction-defined step, pre-treatment (Pre-processing) step (1), the derivative (Derivative) calculation step, the slope representative value (Magnitude) calculation step (2), the contour refinement (edge thinning) step (3), edge detection (detection) step (4), the post-processing (post - processing) step comprises a 5 .

The color definition step, a step to set the multi-color by the plurality of colors constituting the color image and an infrared image received from the multi-color sensor, the direction defined steps, including the x, y direction and the diagonal direction of the image It is a step of determining multi-directionality by a plurality of directions.

The pre-processing step (1) includes a weighted kernel (Gaussian distribution) applied to the color image (RGB) and the infrared image (IR) input from the multiple color sensor, Is applied as the frequency of the weighted median filter. Here, the noise refers to a quantization noise generated in a process of converting from analog to digital. Here, the boundary means a frequency change portion due to the brightness change in the image. Noise and boundary have been confirmed to be sensitive to contour information, and have robustness that improves contour accuracy when the boundary is emphasized.

For this preprocessing, a weighted median filter is used in the present invention. Prior to the explanation, a conventional median filter will be described first.

Fig. 2 shows a process of a conventional median filter.

The conventional median filter derives an intermediate value in consideration of the correlation with neighboring neighboring pixels. This intermediate value efficiently removes noise that may occur on the basis of one pixel and also emphasizes the boundary.

The process of the existing median filter is as follows. The data of {3, 7, 2, 1, 0, 0, 9, 5, 8} 0, 1, 2, 3, 5, 7, 8, 9} is derived and an intermediate value of 3 is selected to replace the original intermediate value (the value of the pixel)

3 shows the process of the weighted median filter of the present invention.

Create a weighting kernel with a Gaussian distribution in advance. When the same data as that of the conventional median filter is input, the input data is ranked as {3, 7, 7, 2, 1, 1 0, 0, 0, 0, 1, 2, 3, 0, 0, 0, 0, 0, 0, 9, 5, 5, 8} 5, 7, 7, 8, 9} is derived, and the median value 2 is selected to replace the original intermediate value (the value of the pixel) with 0.

Here, the weighted kernel is modeled in advance, and when generating the weight, the difference between the spatial distance and the brightness value is considered and is generated based on the window mask. For example, for a 3 x 3 mask, the width of the extended data is 15, and the weighted kernel is {1, 2, 1, 2, 3, 2, 1, 2, 1}.

Fig. 4 shows the results of comparison with a median filter, a weighted median filter, and other conventional filters.

The red circle indicates that the input data weighted median filter maintains the function of noise cancellation and emphasizes the boundary compared to the conventional simple median filter or other filters.

The derivative (Derivative) calculation step, for each color of a multi-color image fixed, and calculating the derivative of each direction in a given multi-directional, the slope representative value (Magnitude) calculating step (2), the obtained derivative And obtaining a slope representative value in the image. It is processed for each image data (RGB + IR) input from the multiple color sensor. Here, the multi-color sensor means a sensor that calculates four colors of R, G, B, and IR.

5 shows a multi-directional model and a derivative operation process.

The multi-directionality is defined as four directions of X, Y, and X45 (45 degrees counterclockwise rotation about X axis in FIG. 5) and Y45 (45 degrees counterclockwise rotation about Y axis in FIG. 5). For example, as shown in FIG. 5, when a 3 × 3 mask is sequentially numbered in the clockwise direction with the upper left cell as No. 1, the derivative of X is a difference value between 4 pixels and 8 pixels, Y The derivative of X45 is the difference value between 3 pixels and 7 pixels, and the derivative of Y45 is the difference value between 5 pixels and 1 pixel.

This derivative process is performed for each of the multiple color sensors (R, G, B, IR) and results in four sensors x 4 directional = 16 derivatives. In order to represent the inclination tendency in the multi-color multi-directionality in combination, it is necessary to use the Euclidean norm formula for changing the frequency to exaggerate the derivative so that, for example, a larger value is emphasized and a smaller value is exaggerated To generate a slope representative value in the image.

Figure 6 shows the definition of Euclidean norm and the process for processing negative numbers. The original definition is the square root of the sum of squares, but the sum of absolute values can be used as an approximation. That is, all the absolute values of the respective derivatives are added together to generate a slope representative value.

Here, a 1-bit amplification process considering a negative number can be applied. There are two cases when amplifying one bit in consideration of a negative number and calculating an absolute value. (256-derivative) is performed on the basis of the 256 value when the absolute value is generated, and when it is larger (256-derivative) when the absolute value is generated, and on the opposite (Derivative-256).

Fig. 7 shows the results of multi-color derivatives taking multi-directionality and the results of slope representative magnitudes.

The contour refinement (edge thinning) Step (3) is, for contour area with a similar pixels together, to obtain an extreme value of the outline from the derivative of the slope representative values, to the outline of the non-peak removal (non-maximum suppression) outlines . It is processed in a conditional form, based on the slope representative value in the image and the derivative value for the multi-directionality. Here, the condition is expressed in eight directions in consideration of the directionality of X and Y. [

Fig. 8 defines eight conditions (cases) for contour refinement and a calculation method therefor. Here, it is assumed that the derivative (G), X increases to the right, and Y increases to the upper side.

(1) Based on the X directionality, case 0xx is defined as a negative part, and case 1xx is defined as a positive part.

(2) With respect to Y direction, case x0x is defined as a negative part, and case x1x is defined as a positive part.

(3) The case xx0 is defined as a negative part, and the case xx1 is defined as a positive part, based on the direction rotated by 45 degrees.

Based on this definition, a case is selected taking into consideration the derivative of the multi-directionality, and a difference value from the slope representative value in the image is extracted to calculate the maximum value. The slope representative values excluding these extremes are removed, resulting in finer results.

Fig. 9 shows the processing before and after processing for contour refinement.

Here, it is understood that refinement of the slope representative value in which similar pixels are gathered improves the accuracy of the contour line.

The contour detection step (4) is a step of determining and detecting contours using adaptive multiple thresholds computed based on in-image histograms. This creates a histogram in the image and adaptively determines the threshold value using the generated histogram. Here, the adaptive threshold value is effective for improving the accuracy of the contour in response to the variable environmental change.

Figure 10 shows the definition of various contours (Strong, weak, necessary) and the need for multiple thresholds.

Within the detailed contour information, the outline is classified into three types: strong contour, Weak contour, and Necessary contour. The strong contour line indicates contour information larger than the high threshold, and the contour line indicates contour information smaller than the high threshold value. And, the four outline contour lines are the outline of the week, but the contour line information adjacent to the strong outline among them.

For their classification, a dual threshold is applied, which is expressed as a high threshold and a low threshold. This is an experimental value. In order to extract four contour lines, it is necessary to design a LIFO (Last Input First Output) method using a recursive method or a stack memory.

11 shows a processing procedure of contour detection.

Create a histogram in the detailed contour image and calculate the total number of pixels in the contour in the histogram. Then, a predetermined ratio of the total number of pixels of the outline is calculated as a threshold value, and a high threshold value and a low threshold value adaptive to the histogram are calculated.

Here, as an experimental dual threshold, for example, a parameter (? _H ) for a high threshold value is set to 0.8, and a parameter (? _H ) for a low threshold value is set to 0.5. Then, the high threshold value is a brightness value corresponding to x (the number of all pixels of the contour on the histogram) x (the parameter (0.8) for the high threshold value), and the low threshold value is the [high threshold value] x (0.5)]. Using this dual threshold, three types of contours (Strong, Weak, Necessary) are selected.

Here, it is desirable to design two types of memory based on memory usage and processing speed considering hardware. In the case of designing with a single threshold, it is designed as a binary with only a low threshold, and 2) when designing with a dual threshold, a high threshold and a low threshold are applied, .

That is, a threshold determination step of determining an adaptive multiple threshold value computed based on a histogram in an image to divide the outline into three types of Strong, Wark, and Four Segments; applying a high threshold value and a low threshold value in the determined threshold value The outline can be detected by the outline detection step of determining and detecting the outline in a recursive manner.

12 shows a timing chart for generating a histogram in the contour detection step.

There is an inefficient aspect that the memory usage becomes vast when the one frame memory is used in the contour detection processing. When analyzing the required memory usage in the process of performing contour detection, it is necessary to use a line buffer (window mask size - 1) in case of processing with a window mask, and compared with the case of using one frame memory .

Here, when the frame rate is 24 frames or more, it is assumed that the image difference between (t-1) and (t) rarely occurs. Based on the design of the line memory type, the number of all pixels of the histogram and the contour line is generated in the previous frame t-1 and applied in the current frame t. That is, the histogram used to calculate the threshold value is the histogram of the previous frame of the frame that calculates the threshold value. Here, the previous frame is defined as (t-1) frame which is one frame before, the current frame is currently (t), and the next frame is defined as (t + 1) which is one frame later.

In FIG. 12, (a) and (b) are synchronized with the active state of Vsync, (a) denotes Rising, and (b) denotes Falling. (b) is defined as the blank interval Vsync (blank) of the Vsync until the falling of (b) occurs and the next rising of (a) occurs. In this interval, the dual threshold is calculated in (c) To the next frame (t + 1).

Figure 13 shows the comparison results after applying the single threshold and multiple thresholds in comparison with the conventional results. The result of applying the double threshold shows that the accuracy of the contour is improved.

The post-processing step (5) is a step of correcting and removing an incorrect part in the detected contour information. Here, the inaccurate part is expressed as "Prejection", "Discontinuous", and "Noise".

Fig. 14 shows definitions of features (protrusion, noise, continuation) to be contour-corrected.

The protrusion refers to a part protruding on a continuous contour, a continuous non-continuous contour information, and a noise part detected as an in-screen noise phenomenon. The size of the protrusion, the perpetuation, and the noise is proportional to the size of the mask, and when the size of the window mask is large, it is possible to correct for the large protrusion and the long continuation.

Fig. 15 shows conditions and processing procedures for erosion and dilation.

Define erosion and swelling as a process for correcting for protrusions, perpetuities, and noises to be corrected. Erosion is the processing of the effect of reducing the data, and expansion is the processing of the effect of expanding the data. Calibration is done in two forms, erosion and filling, using erosion and expansion.

Removal is the role of eliminating protrusion and noise, which can be done by erosion and then by expansion. Filling is a continuing role, which can be accomplished by expansion followed by erosion.

That is, in the post-treatment step, a removal treatment for expanding after erosion to eliminate protrusion and noise and a filling treatment for erosion after expansion are carried out in order to keep continuing.

The size of the kernel is 2N-1 (N is possible from 2), that is, 3 or more odd numbers. Here, Pc denotes a center point in the mask, data 1 denotes an edge, and data 0 denotes a non-edge.

Erosion and expansion include conditions for performance, and the process is also applied differently.

The erosion is applied when Pc corresponds to the contour line (1), that is, the contour line, and the number of pixels of the contour line in the mask is equal to or larger than (the mask width). If the Pc is greater than or equal to [the maximum value x 0.5] in the two-dimensional histogram, the calculation process is preserved (unchanged).

Expansion is applied when Pc is outline (0), that is, when it does not correspond to the outline, and when the number of pixels of the outline in the mask is larger than (mask width - 1). If the Pc in the two-dimensional histogram is maximized, it is generated. Otherwise, the path is unchanged.

Fig. 16 defines the characteristics of the outline to be extracted. The characteristics of the outline to be extracted are divided into a line (first shape), a corner (second shape), and a ridge (curve, third shape).

17 shows a maximum value extraction process based on a two-dimensional histogram.

(X1, X2, X3) (Y1, Y2, Y3) in the X-axis and Y-axis directions and processes it as a product of the matrix operation. A two-dimensional histogram is generated. The maximum value is extracted and compared with Pc.

18 shows an example of a processing procedure for contour correction.

An example of inaccurate outlines is shown in three forms (protrusion, perpetuation, and noise). Indicates that the image is corrected in two steps. The two stages are 1) erosion and 2) expansion.

In the first step, 1) the process of erosion is judged to be inaccurate in the part represented by 1 where the contour exists, and if it is judged to be inaccurate, it is expressed in shade. As a result, the shaded part is treated as an outline of zero.

In the second process, 2) the process of expansion is examined, the inaccurate outline is judged in the part represented by 0 where there is no contour line, and if it is judged to be inaccurate, it is expressed in shaded color. As a result, the shaded part is treated as a single outline.

The cross-shaped protrusions, the independent small points, and the perturbations are corrected to produce a linear-shaped result.

Fig. 19 shows the results for contour correction.

Performing erosion (erosion), which is a one-step process on the detected contour, removes the point of protrusion and noise. Thereafter, when the expansion (expansion) which is the second process is performed, a point is generated at the continuous portion to maintain the continuity. The red circle shows erosion and expansion to compensate for such phenomena as protrusion, perpetuation, and noise.

Although the present invention has been described with reference to specific embodiments, the present invention is not limited thereto. Modifications, modifications and variations made by those skilled in the art within the scope of the claims are to be construed as being within the scope of the present invention.

INDUSTRIAL APPLICABILITY The present invention can be applied to an apparatus and method for extracting contours of an image based on multiple colors and multiple directions.

1: Pre-processing unit
2: Derivative and slope representative value extraction unit
3: edge thinning
4: contour detection unit
5: Post-processing unit

Claims (25)

A method of extracting contours of an image based on multiple colors and multi-directionality,
A color definition step of defining multiple colors by a plurality of colors constituting a color image and an infrared image received from a multi-color sensor,
A direction defining step of defining multiple directions by a plurality of directions including x, y and diagonal directions of an image;
A derivative calculation step of obtaining, for an image of each color of a predetermined multiple color, a derivative of each direction of a predetermined multi-directionality;
A slope representative value calculating step of calculating a derived derivative and calculating a slope representative value in the image;
An outline refinement step of obtaining an extreme outline from a derivative and a slope representative value for an outline area in which similar pixels are gathered and removing an outline other than extrema to thereby reduce the outline;
An outline detection step of determining and detecting an outline using an adaptive multiple threshold computed based on an in-image histogram;
And extracting a contour of an image based on multiple colors and multiple directions. The method according to claim 1,
The derivative calculating step before, for the color image and an infrared image received from the multi-color sensor, a pre-processing to remove noise and highlight the boundary (Pre - processing) step or,
After the outline detection step, the protrusion, the continuation, and the noise in the detection result are removed After processing (Post - processing) step
And extracting a contour of an image based on multiple colors and multiple directions. The method according to claim 1 or 2,
The arithmetic processing for obtaining the slope representative value is performed by using the Euclidean gamma
A method for extracting contours of an image based on multiple colors and multi-directionality. The method according to claim 1 or 2,
The arithmetic processing for obtaining the slope representative value is a square root processing of the sum of squares of respective derivatives.
A method for extracting contours of an image based on multiple colors and multi-directionality. The method according to claim 1 or 2,
The arithmetic processing for obtaining the slope representative value is a summing process of absolute values of respective derivatives
A method for extracting contours of an image based on multiple colors and multi-directionality. The method according to claim 1 or 2,
In the contour refinement step, considering the x and y directions
(1) Based on the X directionality, case 0xx is defined as a negative part, case 1xx is defined as a positive part,
(2) With respect to the Y direction, the case x0x is defined as a negative part, and the case x1x is defined as a positive part,
(3) Based on the direction rotated by 45 degrees, case xx0 is defined as a negative part, case xx1 is defined as a positive part,
8-direction case,
Based on the above definition, the case is selected in consideration of the derivative of the multi-directionality, the difference value from the slope representative value in the image is extracted, and the extremal value is calculated
A method for extracting contours of an image based on multiple colors and multi-directionality. The method according to claim 1 or 2,
In the contour detection step, the histogram is generated, the total number of pixels of the contour is calculated, and a predetermined ratio of the total number of pixels of the contour is calculated as a threshold value
A method for extracting contours of an image based on multiple colors and multi-directionality. The method according to claim 1 or 2,
The histogram is a histogram of a previous frame of the frame for which the threshold value is to be calculated
A method for extracting contours of an image based on multiple colors and multi-directionality. The method of claim 2,
In the preprocessing step, a weighted median filter is applied in which a weighted kernel to which a Gaussian distribution is applied is applied as a frequency.
A method for extracting contours of an image based on multiple colors and multi-directionality. The method of claim 2,
In the post-treatment step, the post-erosion expansion process is performed to remove the protrusion and the noise, and the post-expansion erosion process is carried out to perform the filling process
A method for extracting contours of an image based on multiple colors and multi-directionality. An apparatus for extracting contours of an image based on multiple colors and multiple directions,
A color definition unit for determining multiple colors by a plurality of colors constituting a color image and an infrared image received from the multiple color sensor,
A direction defining unit for defining multiple directions by a plurality of directions including x, y and diagonal directions of the image;
A derivative calculator for obtaining a derivative with respect to each direction of a predetermined multi-directionality for an image of each color of a predetermined multi-color,
A slope representative value calculating unit for calculating a derived derivative and calculating a slope representative value in the image,
An outline refinement section for obtaining outlines of extreme values from a derivative and a slope representative value for a contour area in which similar pixels are gathered and removing contours other than the extrema,
An outline detector for determining and detecting an outline by using an adaptive multiple threshold computed based on an in-image histogram,
And an image extracting unit for extracting an outline of an image based on multiple colors and multiple directions. The method according to claim 1,
Before the derivative calculation unit operation, for the color image and an infrared image received from the multi-color sensor, a pre-processing to remove noise and highlight the boundary (Pre - processing) unit or,
After the contour detection step operation, the protrusion, the perception, and the noise in the detection result are removed The post-processing unit
Further comprising: an image extraction unit for extracting contours of an image based on multiple colors and multiple directions. The method according to claim 1 or 2,
The arithmetic processing for obtaining the slope representative value is performed by using the Euclidean gamma
And extracting contours of images based on multiple colors and multiple directions. The method according to claim 1 or 2,
The arithmetic processing for obtaining the slope representative value is a square root processing of the sum of squares of respective derivatives.
And extracting contours of images based on multiple colors and multiple directions. The method according to claim 1 or 2,
The arithmetic processing for obtaining the slope representative value is a summing process of absolute values of respective derivatives
And extracting contours of images based on multiple colors and multiple directions. The method according to claim 1 or 2,
In the contour refinement step, considering the x and y directions
(1) Based on the X directionality, case 0xx is defined as a negative part, case 1xx is defined as a positive part,
(2) With respect to the Y direction, the case x0x is defined as a negative part, and the case x1x is defined as a positive part,
(3) Based on the direction rotated by 45 degrees, case xx0 is defined as a negative part, case xx1 is defined as a positive part,
8-direction case,
Based on the above definition, the case is selected in consideration of the derivative of the multi-directionality, the difference value from the slope representative value in the image is extracted, and the extremal value is calculated
And extracting contours of images based on multiple colors and multiple directions. The method according to claim 1 or 2,
In the contour detection step, the histogram is generated, the total number of pixels of the contour is calculated, and a predetermined ratio of the total number of pixels of the contour is calculated as a threshold value
And extracting contours of images based on multiple colors and multiple directions. The method according to claim 1 or 2,
The histogram is a histogram of a previous frame of the frame for which the threshold value is to be calculated
And extracting contours of images based on multiple colors and multiple directions. The method of claim 2,
In the preprocessing step, a weighted median filter is applied in which a weighted kernel to which a Gaussian distribution is applied is applied as a frequency.
And extracting contours of images based on multiple colors and multiple directions. The method of claim 2,
In the post-treatment step, the post-erosion expansion process is performed to remove the protrusion and the noise, and the post-expansion erosion process is carried out to perform the filling process
And extracting contours of images based on multiple colors and multiple directions. In order to extract the contours of an image based on multiple colors and multi-directionality,
A color definition step of defining multiple colors by a plurality of colors constituting a color image and an infrared image received from a multi-color sensor,
A direction defining step of defining multiple directions by a plurality of directions including x, y and diagonal directions of an image;
A derivative calculation step of obtaining, for an image of each color of a predetermined multiple color, a derivative of each direction of a predetermined multi-directionality;
A slope representative value calculating step of calculating a derived derivative and calculating a slope representative value in the image;
An outline refinement step of obtaining an extreme outline from a derivative and a slope representative value for an outline area in which similar pixels are gathered and removing an outline other than extrema to thereby reduce the outline;
An outline detection step of determining and detecting an outline using an adaptive multiple threshold computed based on an in-image histogram;
Readable medium having recorded thereon a program for executing a computer readable medium. A method for detecting a contour of an image,
A threshold determination step of determining an adaptive multiple threshold computed based on an in-image histogram in order to classify the outline into three types of strong, wek, and four sets;
An outline detection step of determining and detecting an outline in a recursive manner by applying a high threshold value and a low threshold value among the determined thresholds
And detecting a contour of the image. As a method for removing protrusion, continuation, and noise in a contour extracted from an image,
A removal treatment step of performing expansion after erosion to eliminate protrusion and noise,
A filling treatment step for carrying out erosion after expansion to maintain continuity
And extracting a contour line from the image. 24. The method of claim 23,
The erosion is performed when a contour line exists in the center pixel of the mask and the number of pixels of the contour line in the mask is equal to or larger than [the mask width]
The erosion is unchanged when the center pixel of the mask in the two-dimensional histogram is greater than or equal to [maximum value x 0.5], and is removed when it is small
Wherein the method comprises the steps of: extracting a contour line from a video image; 24. The method of claim 23,
Preferably,
Processing is performed when there is no outline in the center pixel of the mask and the number of pixels of the outline in the mask is equal to or larger than [mask width - 1]
When the center pixel of the mask in the two-dimensional histogram reaches the maximum, it is generated.
Wherein the method comprises the steps of: extracting a contour line from a video image;

Images