KR102282086B1 - Apparatus For Generating Disparity Map Using Composite Filter And Thereof Method - Google Patents

Abstract

The present invention is a lighting unit for irradiating light to an object using a plurality of lamps having different irradiation angles; a sensor unit which acquires images with different parallaxes with an image sensor that scans the surface of the object on which the light of the illumination unit is illuminated; The images with different parallax obtained by the sensor unit are input, the kernel function setter sets a synthetic kernel function by combining kernel functions independently applied in filter units, and the parallax difference is obtained using the synthetic filter to which the synthetic kernel function is applied. an image processing unit configured to perform a filtering operation on different images, and matching the filtered images with different parallaxes to generate a disparity image, wherein the kernel function setter includes different kernels for at least one filter parameter that determines a kernel characteristic. It is characterized in that the composite kernel function is set by combining the functions.

Description

Apparatus For Generating Disparity Map Using Composite Filter And Thereof Method

The present invention relates to an apparatus and method for generating a disparity image using a synthesis filter capable of extracting feature points of a stereo image by setting a characteristic function of a filter according to a target object and providing a disparity image based thereon.

A method of obtaining a depth map in stereo vision is to obtain a depth image by similarly applying binocular disparity generated from both eyes when a person sees an object in a stereo camera. At this time, the binocular disparity in the stereo camera is called disparity, and the process of finding the disparity is obtained by finding a matching point in two images of the same scene.

As a prior art, [Patent Document 1] discloses a technique for obtaining a depth image using a window-based sum of absolute difference (SAD). Using this method, depth information of an image can be extracted, but when applied to a real-time image, there is a problem in that a lot of noise is generated.

[Patent Document 1] Korean Patent No. 10-1988551 (Announcement on June 12, 2019)

does not exist

An object of the present invention is to provide an apparatus and method for generating a disparity image using a synthesis filter capable of obtaining a disparity image from a stereo image.

In order to achieve the above object, an apparatus for generating a disparity image using a synthesis filter according to the present invention includes: an illumination unit irradiating light to an object using a plurality of lamps having different irradiation angles; a sensor unit which acquires images with different parallaxes with an image sensor that scans the surface of the object on which the light of the illumination unit is illuminated; The images with different parallax obtained by the sensor unit are input, the kernel function setter sets a synthetic kernel function by combining kernel functions independently applied in filter units, and the parallax difference is obtained using the synthetic filter to which the synthetic kernel function is applied. an image processing unit configured to perform a filtering operation on different images, and matching the filtered images with different parallaxes to generate a disparity image, wherein the kernel function setter includes different kernels for at least one filter parameter that determines a kernel characteristic. It is characterized in that the composite kernel function is set by combining the functions.

In addition, the synthesis filter applies a Gabor filter, and the kernel function setter includes a kernel function averaging unit for arithmetic average of kernel functions obtained by changing a directional parameter as a filter parameter defining the kernel function of the Gabor filter. do.

In addition, the kernel function setter individually outputs the design values of the corresponding directional parameters using a filter information storage unit for storing filter information necessary for setting the synthetic kernel function, and a plurality of registers for storing design values of different directional parameters. A filter parameter output unit, a heirloom kernel function output unit for outputting the corresponding heirloom kernel function to the kernel function averaging unit using a plurality of heirloom kernel function operators for calculating a heirloom kernel function using the filter parameters of the heirloom filter, and the filter parameter A switching unit having a plurality of switches connected between the output unit and the heirloom kernel function output unit and individually providing a corresponding directional parameter design value to a corresponding heirloom kernel function calculator according to an operation signal, and filter parameter identification information from the filter information storage unit It is characterized in that it comprises a filter parameter selection unit for outputting an operation signal for individually operating the switch of the switching unit by receiving the.

In addition, the kernel function averaging unit is characterized in that, when arithmetic average is performed according to the rule mode, the average kernel function (g _avr-n ) is obtained by using the following equation.

Here, x and y are the coordinate values in the rectangular coordinate system of the target image, λ is the wavelength that controls the sine function of the Gabor filter kernel, θ is the directionality of the kernel function, ψ is the phase difference, σ is the standard deviation of the Gaussian function, and γ is the Gabor function. It is a factor that determines the shape of the filter, and n is the number of division for arithmetic mean.

In addition, when the kernel function averaging unit uses three kernel functions according to the rule mode, three directional parameters (θ ₁ , θ ₂ , θ ₃ ) corresponding to each kernel function are 0, π/3, and 2π/3. characterized in that

In addition, the kernel function averaging unit is characterized in that _{the average kernel function (g avr-n} ) is obtained by using the following equation when arithmetic average is performed according to the irregular mode.

Here, x and y are the coordinate values in the rectangular coordinate system of the target image, λ is the wavelength that controls the sine function of the Gabor filter kernel, θ is the directionality of the kernel function, ψ is the phase difference, σ is the standard deviation of the Gaussian function, and γ is the Gabor function. It is a factor that determines the shape of the filter, and n+1 is the number of division for arithmetic mean.

In addition, the lighting unit is characterized in that it is provided with four lamps disposed on the upper portion of the object.

In addition, the image processing unit is disposed above the object and is connected to one side of first and second image sensors for obtaining images with different parallaxes, and includes a filter unit having first and second synthesis filters for filtering operation; and a stereo image matching unit outputting one image by matching the stereo image output from the second synthesis filter, and a disparity image generating unit generating a disparity image using the image matched by the stereo image matching unit. characterized in that

Disparity image generation method using a synthesis filter according to the present invention for achieving the above object is an image with different parallax by illuminating the surface of an object using a plurality of lamps having different irradiation angles and scanning the surface of the object with an image sensor obtaining them; A synthesis filter receives the images obtained in the image acquisition step and performs filtering operations, but the synthesis kernel function applied to the synthesis filter is set by combining different kernel functions by a kernel function setter for at least one filter parameter that determines kernel characteristics and outputting a filtered stereo image; calculating disparity information existing between one image and another image by matching the filtered stereo image in the synthesis filter; and generating a disparity image using disparity information obtained when the stereo image is matched.

In addition, when the synthesis filter applies the Gabor filtering technique, it is characterized in that the synthesis kernel function of the synthesis filter is set by arithmetic average of kernel functions defined by the directionality parameter that determines the directionality of the Gabor filter.

In addition, when setting the arithmetic mean division number for dividing the combined kernel function when setting the synthetic kernel function, the same number as the number specified by the user is used according to the rule mode or the number specified by the user according to the irregular mode It is characterized by using a larger number than a predetermined number.

As described above, the present invention can provide a high-quality disparity image in which the disparity information calculated based on the extracted feature points of the stereo image is extracted by setting the characteristic function of the synthesis filter according to the target object.

In addition, according to the present invention, a synthesis kernel function of a synthesis filter is set by combining kernel functions having different design values of filter parameters, and the amount of calculation required for a filtering operation can be significantly reduced by using the synthesis filter.

1 is a schematic diagram illustrating an operation of acquiring a stereo image using two image sensors and four lamps disposed on an object to generate a disparity image of a target object according to the present invention;
2 is a block diagram of an apparatus for generating a disparity image using a synthesis filter according to the present invention;
3 is a detailed block diagram of the kernel function setter of FIG. 2;
4 is a view showing a shape of a kernel that varies depending on a filter parameter that determines the kernel characteristic of a Gabor filter;
5 is a test image showing a real image viewed from the top of a printed circuit board and a part of a red box to the right for performance inspection of a synthetic filter according to the present invention;
6 is an output image showing the result of filtering operation by applying two kernels having different directional parameters to the sample image 1, which is a part of the test image of FIG. 5;
7 is a view exemplarily showing each kernel to be applied to the synthesis filter according to the present invention;
8 is a view showing heirloom kernel 1, heirloom kernel 3, and heirloom kernel 5 having different kernel characteristics from sample image 2, which is a part of the test image of FIG. 5;
9 is an output image showing the result of filtering operation by individually applying the real part, the imaginary part, and the whole of each kernel when heirloom kernel 1, heirloom kernel 3, and heirloom kernel 5 are applied to sample image 2;
10 is an enlarged image of an image output from the first and second synthesis filters to which the imaginary part of the heirloom kernel 3 is applied and a part of the image;
11 is a disparity image to which the conventional SAD technique is applied to the test image of FIG. 5, a disparity image to which a synthesis filter according to the present invention is applied, and an enlarged image of a portion of each disparity image.

Hereinafter, the present invention will be described by describing embodiments of the present invention with reference to the accompanying drawings.

Like reference numerals in each figure indicate like elements. Also, in describing the present invention, if it is determined that a detailed description of a related known function or configuration may unnecessarily obscure the gist of the present invention, the detailed description thereof will be omitted. In addition, when a part "includes" a certain component, this means that other components may be further included, rather than excluding other components, unless otherwise stated.

In the field of verifying the identity of a visitor by recognizing a person's face and vein pattern, etc., in the field of inspecting the surface defects of various electronic products such as printed circuit boards, display panels and secondary batteries, etc. Stereo vision systems are being used in various technical fields such as inspection fields.

The present invention can be applied to a stereo vision system that obtains a three-dimensional image by image processing a stereo image. Stereo image matching is a technique used to obtain a three-dimensional image from a stereo image, and is used to obtain a three-dimensional image from a plurality of two-dimensional images taken at different photographing positions on the same line with respect to the same object. In this case, a filter is used for pre-processing the stereo image. For example, the 2D Gabor filter is a filter that can freely pass only components of a specific position, a specific frequency, and a specific direction with respect to a target signal. Techniques for image processing in the spatial domain by utilizing the advantages of such a heirloom filter, for example, image enhancement, image restoration, image texture analysis, and Gabor filters are widely used in image segmentation.

Gabor filters have filter parameters related to tunable orientation and center frequency, and radial frequency bandwidth in spatial and frequency domains. Therefore, depending on how the filter parameters are set, the shape, orientation, center frequency, bandwidth, etc. of the Gabor filter are changed.

When the heuristic filtering technique is used for image processing, after all filters with different frequencies, directions, and bandwidths are applied to the input image, a spatial filter showing the optimal response among them can be used. When using a multi-channel bank, there is a problem in that the amount of calculation is very large because it is complicated and requires a large amount of calculation.

In order to solve the problem of the prior art of applying one filter with the best responsiveness among them by looking at the results of applying one by one to a plurality of heirloom filters, the present invention sets a filter parameter that determines the filter characteristics according to the target object. A synthesis filter is used as an improved Gabor filter that can increase the accuracy of extraction and significantly reduce the amount of computation.

1 is a schematic diagram illustrating an operation of acquiring a stereo image using two image sensors and four lamps disposed on an object to generate a disparity image of a target object according to the present invention, and FIG. It is a block diagram of an apparatus for generating a disparity image using a synthesis filter according to the present invention.

In the apparatus for generating disparity using a synthesis filter according to the present invention, the lighting unit 20 and the sensor unit 10 are disposed on the target object P. Here, the target object P may be fixed to a jig (not shown) or slidably moved by a moving table depending on the use environment, but the target object P does not need to be specified. For example, not only various electronic products for inspecting surface defects through image analysis, but also a part of a human body such as a face or a vein pattern for identification may be a target.

The sensor unit 10 includes image sensors 11 and 12 to obtain a stereo image of the surface of the object P. The two image sensors 11 and 12 are installed at regular intervals on the same line to obtain stereo images with different parallax, and may be implemented as, for example, a charge coupled device (CCD).

The lighting unit 20 is for illuminating the surface of the object P, and includes a plurality of lamps 21 , 22 , 23 , and 24 having different irradiation angles. The four lamps 21 , 22 , 23 , and 24 may be radially disposed on the object so that light is incident from all directions toward the object P to illuminate the entire area without a shaded area, for example, light emission. It may be implemented as a light-emitting diode (LED).

As shown in FIG. 2 , in the apparatus for generating a disparity image using a synthesis filter according to the present invention, the image processing unit 30 includes an illumination unit 10 , a sensor unit 20 , an input unit 80 , and a monitor 90 . can be electrically connected.

In an environment where all four lamps (21, 22, 23, 24) are turned on and light is shone on the surface of the object (P), the two image sensors (11, 12) scan the surface of the object (P) and Acquire an image. That is, the two image sensors 11 and 12 convert the light entering through the lens into a digital signal to form a stereo image, and the other image paired with the first image acquired from one image sensor 11 The second image acquired from the sensor 12 constitutes a stereo image acquired from the same object P. The obtained stereo image is output to the image processing unit 30 .

The image processing unit 30 includes a kernel function setter 40 , a filter unit 50 , a stereo image matching unit 60 , and a disparity image generation unit 70 .

The filter unit 50 extracts the feature points of the stereo image.

The filter unit 50 includes first and second synthesis filters 51 and 52 for performing a filtering operation on the stereo images received from the first and second image sensors 11 and 12, respectively. The first synthesis filter 51 performs a convolution operation on all pixels of the first image obtained by the first image sensor 11 , and the second synthesis filter 52 performs a convolution operation on the first image obtained by the second image sensor 12 . A convolution operation is performed on all pixels of the two images, and after each filtering operation is performed by the first and second synthesis filters 51 in this way, the feature points of the image are extracted and then input to the stereo image matching unit 60 .

A Gabor filter may be applied to the first and second synthesis filters 51 and 52 of the filter unit 50 . When the feature point extraction of the image is faithfully performed by the filter unit 50 , the stereo image matching unit 60 may obtain accurate disparity information. Since the filter characteristics of the heirloom filter are determined by the filter parameters of the heirloom kernel function, when designing the filter for the filter unit 50, the setting of the kernel function that determines the filter characteristics of the filter unit 50 should be considered important. In the embodiment, each synthesis kernel function for the first and second synthesis filters 51 and 52 is set using a kernel function setter 40 to be described later.

The stereo image matching unit 60 calculates disparity information by matching the stereo images that have undergone the filtering operation in the first and second synthesis filters 51 and 52, and The pixel and the pixel of the second image input from the second synthesis filter 52 are compared to find a point of agreement according to the degree of similarity between the pixels, and the distance between the two pixels is obtained as a parallax.

The disparity image generating unit 70 generates a disparity image expressing a sense of depth in black and white based on the disparity information calculated by the stereo image matching unit 60 . This disparity image is output to the monitor 90 and can be visually confirmed.

3, the synthesis kernel functions applied to the first and second synthesis filters 51 and 52 are respectively set by the kernel function setter 40, and the same synthesis kernel function is set. A kernel function setter 40 receiving a user command from the input unit 80 sets a synthesis kernel function applied to the first and second synthesis filters 51 and 52, and a synthesis kernel function according to the target object P. can be set differently. In the embodiment, a Gabor filter is employed as the first and second synthesis filters 51 and 52, and a directional parameter is selected as a filter parameter that determines the filter characteristics of the Gabor filter. That is, the kernel function setter 40 sets the synthesized kernel function by changing at least one filter parameter that determines the filter characteristics.

Unlike the embodiment, the synthesis kernel function may be set by targeting filter parameters other than the directional parameter. Also, when an image filter other than the heirloom filter is applied, a synthesis kernel function can be set by combining the filter parameters of the image filter.

The kernel function setter 40 includes a filter parameter selection unit 41, a filter parameter output unit 42, a switching unit S1, S2, ..., Sn, a heirloom kernel function output unit 43, a kernel function average operation unit ( 44) and a filter information storage unit 45 .

The filter information storage unit 45 provides filter information applied to the filter unit 50 according to a user command of the input unit 80 to the filter parameter selection unit 41 and the kernel function average operation unit 44 . Here, the filter information means identification information of a filter parameter selected to set the synthesis kernel function, and the number of divisions of the arithmetic mean. The filter parameter identification information is provided to the filter parameter selection unit 41 , and the division number of the arithmetic mean is provided to the kernel function average operation unit 44 .

The filter parameter output unit 42 includes first to n-th registers h1, h2, ..., hn for individually outputting design values for directional parameters. The first register h1 stores the first directional parameter θ ₁ , the second register h2 stores the second directional parameter θ ₂ , and the nth register hn stores the nth directional parameter θ 2 . θ _n ) can be stored.

The switching units S1, S2, ..., Sn, which operate according to the operation signal of the filter parameter selection unit 41, are connected one-to-one to the output side of the filter parameter output unit 42 . That is, a corresponding one of the first to n-th switches S1, S2, ..., Sn is connected in series to the output side of the first to n-th registers h1, h2, ..., hn.

The Gabor kernel function output unit 43 is connected to one side of the first to n-th switches S1, S2, ..., Sn. The heirloom kernel function output unit 43 includes first to n-th operators F1, F2, ..., Fn for individually outputting calculated values for the heirloom kernel function. The first operator F1 calculates and outputs the first heirloom kernel function g1 to which the first directional parameter design value θ ₁ received through the first switch S1 is applied, and the second operator F2 outputs the second The second heirloom kernel function g2 to which the second directional parameter design value θ ₂ received through the second switch S2 is applied is calculated and output, and the nth operator Fn is received through the nth switch Sn The n-th heirloom kernel function gn to which the n-th directional parameter design value θ _{n is applied is calculated and output.}

The kernel function averaging unit 44 receives the first to nth directional parameter design values θ ₁ , θ ₂ , ..., θ _n provided from the heirloom kernel function output unit 43 and the filter information storage unit 45 . The average operation is performed using the arithmetic mean division number, and the average kernel function g _avr-n obtained as a result of the operation is provided to the first and second synthesis filters 51 and 52 .

Here, the number of divisions of the arithmetic mean is set differently according to the rule mode and the irregular mode. That is, the rule mode uses a combination of the same number of kernel functions as the designated number (n) input by the user, and the irregular mode uses a larger number than the designated number (n) input by the user (1 is added in the embodiment). )) in combination with the kernel functions. In the embodiment, when the user designates any one of 1 to 4 as the designated number (n), the division number of the arithmetic mean is set according to the rule mode, and when the user designates 5 as the designated number (n) It is predetermined to set the number of divisions of the arithmetic mean according to the irregular mode.

Referring to [Table 1], when the specified number (n) is 4 or less, the number of divisions of the arithmetic mean is the same as the number of kernel functions to be combined, so the same number of design values for the directionality parameters are required. When the designated number (n) is 5, since the number of division of the arithmetic mean is 6 in order to combine the six kernel functions, six directional parameter design values are required. Each direction parameter design value may be set differently according to the target object P.

designated number Filter parameter identification information Directional parameter design value n=1 Directional parameter (θ) θ ₁ =0 n=2 Directional parameter (θ) θ ₁ =0,θ ₂ =π/2 n=3 Directional parameter (θ) θ ₁ =0,θ ₂ =π/3,θ ₃ =2π/3 n=4 Directional parameter (θ) θ ₁ =0,θ ₂ =π/4,θ ₃ =π/2,θ ₄ =3π/4 n=5 Directional parameter (θ) θ ₁ =0,θ ₂ =π/6,θ ₃ =π/3,θ ₄ =π/2,θ ₅ =2π/3,θ ₆ =5π/6

When the kernel function averaging unit 44 arithmetic averages four or less kernel functions according to the rule mode, the average kernel function g _avr-n is obtained using the following [Equation 1].

Here, x and y are the coordinate values in the rectangular coordinate system of the target image, λ is the wavelength that controls the sine function of the Gabor filter kernel, θ is the directionality of the kernel function, ψ is the phase difference, σ is the standard deviation of the Gaussian function, and γ is the Gabor function. It is a factor that determines the shape of the filter, and n is the number of division for arithmetic mean.

For example, a user command for specifying the designated number n from the input unit 80 as 3 and applying the first to third directional parameters (θ ₁ , θ ₂ , θ ₃ ) according to the rule mode is a kernel function setter When input to ( 40 ), the filter parameter selection unit 41 outputs an operation signal for turning on the first to third switches S1 , S2 , and S3 . Accordingly, the first to third heirloom kernel function operators F1, F2, and F3 connected to the first to third switches S1, S2, and S3 operate through the first to third switches S1, S2, and S3. _{The first to third directional parameter design values θ 1} , θ ₂ , and θ ₃ are respectively input from the first to third registers h1 , h2 , and h3 to calculate the heirloom kernel functions g1 , g2 , and g3 .

The first heirloom kernel function operator F1 calculates the first heirloom kernel function g1, the second heirloom kernel function operator F2 calculates the second heirloom kernel function g2, and the third heirloom kernel function operator (F3) calculates the third heirloom kernel function g3.

The kernel function averaging unit 44 divides the three kernel functions g1, g2, and g3 by 3, which is an arithmetic average division number, to obtain an average kernel function g _avr-3 .

Referring to [Table 1], the three directional parameters (θ ₁ , θ ₂ , θ ₃ ) applied to each kernel function may be 0, π/3, and 2π/3.

As another example, the designated number n from the input unit 80 is designated as 5, and the first to sixth direction parameters (θ ₁ , θ ₂ , θ ₃ , θ ₄ , θ ₅ , θ ₆ ) are assigned according to the irregular mode. When a user command for application is input to the kernel function setter 40, the kernel function average operation unit 44 obtains an average kernel function g _avr-n using the following [Equation 2].

Here, x and y are the coordinate values in the rectangular coordinate system of the target image, λ is the wavelength that controls the sine function of the Gabor filter kernel, θ is the directionality of the kernel function, ψ is the phase difference, σ is the standard deviation of the Gaussian function, and γ is the Gabor function. It is a factor that determines the shape of the filter, and n+1 is the number of division for arithmetic mean.

Referring to [Table 1], the six directional parameters (θ ₁ , θ ₂ , θ ₃ , θ ₄ , θ ₅ , θ ₆ ) applied to each kernel function are 0, π/6, π/3, π/ 2, 2π/3, 5π/6 may be used.

As described above, when the respective synthesis kernel functions for the first and second synthesis filters 51 and 52 are set using the kernel function setter 40, the stereo image matching unit 60 performs the first and second synthesis By matching the stereo images that have undergone the filtering operation in the filters 51 and 52 , disparity information is calculated and provided to the disparity image generator 70 . The disparity image generating unit 70 may generate a disparity image capable of expressing a sense of depth in black and white based on the disparity information calculated by the stereo image matching unit 60 . This disparity image is output to the monitor 90 and can be visually confirmed.

Hereinafter, the process of generating a disparity image using the synthesis filter according to the present invention is verified and compared with the prior art, and this will be described.

As shown in FIG. 4 , the 2D Gabor filter can variously change the shape of the kernel by adjusting filter parameters.

5 shows a real image and a test image viewed from the top of a printed circuit board for performance testing of a synthetic filter according to the present invention, and the test image on the right is an enlarged view of a part of a red box in the real image on the right .

As shown in FIG. 6 , when two kernels having different directional parameters are respectively applied to [Sample Image 1], which is the green box area in the test image of FIG. 5, it can be seen that the extraction of key points of the image is different. For example, in [Output Image 1] obtained through filtering operation using a kernel in which the directionality parameter of the Gabor filter is set to π/2, the square corners are prominent, whereas a kernel in which the directionality parameter of the Gabor filter is set to π/4 is used. Thus, it can be seen that ripples appear in [Output Image 2] obtained through the filtering operation.

7 is a view exemplarily showing each kernel to be applied to the synthesis filter according to the present invention. [Heirloom Kernel 1] is a first directionality according to a rule mode in which a designated number (n) is designated as 1 according to a user command. The shape of the kernel for each of the synthesis filters 51 and 52 to which the parameter (θ ₁ ) is applied, and [Heirloom Kernel 2] is the first and second according to the rule mode in which the designated number (n) is designated as 2 according to the user command. This is the shape of the kernel for each synthesis filter 51 and 52 to which the bidirectional parameters (θ ₁ , θ ₂ ) are applied, and [Heirloom Kernel 3] is in the rule mode in which the specified number (n) is specified as 3 according to the user command. The shape of the kernel for each synthesis filter 51 and 52 to which the first to third directional parameters (θ ₁ , θ ₂ , θ _{3 ) are applied according to the} ) is the shape of the kernel for each synthesis filter 51 and 52 to which the first to fourth directional parameters (θ ₁ , θ ₂ , θ ₃ , θ _{4 ) are applied according to the rule mode in which 4 is specified, 5] is each of the first to sixth direction parameters (θ 1} , θ ₂ , θ ₃ , θ ₄ , θ ₅ , θ ₆ ) according to the irregular mode in which the designated number (n) is designated as 5 according to the user command. The shape of the kernel for the synthesis filters 51 and 52 is shown.

8 shows the shapes of [Sample Image 2], which are a part of the test image of FIG. 5, and Heirloom Kernel 1, Heirloom Kernel 3, and Heirloom Kernel 5 having different kernel characteristics in FIG. 7, and as shown in FIG. 9, the same [Sample Image] Image 2], when heirloom kernel 1, heirloom kernel 3, and heirloom kernel 5 are applied, the feature point extraction of the image is different.

[Heirloom Kernel 1-R], [Heirloom Kernel3-R], [Heirloom Kernel 5-R] are different from each other. Also, for the same [Sample Image 2], [Heirloom Kernel 1-I], [Heirloom Kernel 3-I], [Heirloom Kernel 1-I], [Heirloom Kernel 3-I], [Heirloom Kernel 1-I], Kernel 5-I] are different from each other. Also, for the same [Sample Image 2], [Heirloom Kernel 1-C], [Heirloom Kernel3-C], [Heirloom Kernel 1-C], [Heirloom Kernel 3-C], Kernel 5-C] are different.

Also, for the same [Sample Image 2], [Heirloom Kernel 1-R], [Heirloom Kernel 1-I], [Heirloom Kernel 1] obtained by applying the real part, the imaginary part, and the whole of each kernel function to the heirloom kernel 1. -C] is also different, and similarly, images obtained when the real part, imaginary part, and whole of each kernel function are applied to the heirloom kernel 3 and heirloom kernel 5 for the same [sample image 2] are different from each other .

As shown in FIG. 9, the feature point extraction varies according to the kernel function of the Gabor filter, and the disparity image generator obtained from the disparity information matched by the stereo matching unit 60 according to the filtering operation result of the Gabor filter ( 70), the quality of the disparity image is closely affected.

test example

- Synthesis kernel function of the first and second synthesis filters: [Gabor kernel 3-I] applied

- Image size: 370×1030 [pixel]

- Image processing time: 4.5ms

As shown in FIG. 10 , the image output from the first and second synthesis filters 51 and 52 obtained according to the test example has good feature point extraction. The first image St1-k3 filtered by the first synthesis filter 51 with respect to the output image St1 of the first image sensor 11 and the output image St2 of the second image sensor 12 The second image (St2-k3) filtered by the second synthesis filter 52 showed almost similar feature point extraction, and as shown in each enlarged image (k3-E1, k3-E2), the edge boundary portion It can be seen that this stands out prominently.

In this way, the stereo image matching unit 60 matches the filtered images from the first and second synthesis filters 51 and 52 to calculate disparity information, and based on the calculated disparity information, the disparity image generation unit 70 ) generates a disparity image. 11, on the left side, the disparity image [DP-E1] obtained by applying the conventional SAD method and the disparity image [DP-E2] obtained according to the test example using the synthesis filter of the present invention are compared. It can be seen that the disparity image [DP-E2] on the right faithfully reproduces the sense of depth.

The above description of the present invention is for illustration, and those of ordinary skill in the art to which the present invention pertains can understand that it can be easily modified into other specific forms without changing the technical spirit or essential features of the present invention. will be.

10: sensor unit 20: lighting unit
30: image processing unit 40: kernel function setter
41: filter parameter selection unit 42: filter parameter output unit
43: heirloom kernel function output unit 44: kernel function average operation unit
45: filter information storage unit 50: filter unit
60: stereo image matching unit 70: disparity image generating unit
80: input unit 90: monitor

Claims (11)

an illumination unit irradiating light onto an object using a plurality of lamps having different irradiation angles;
a sensor unit which acquires images with different parallaxes with an image sensor that scans the surface of the object on which the light of the illumination unit is irradiated;
The images with different parallax obtained by the sensor unit are input, the kernel function setter sets a synthetic kernel function by combining kernel functions independently applied in filter units, and the parallax difference is obtained using the synthetic filter to which the synthetic kernel function is applied. An image processing unit configured to perform a filtering operation on different images, and matching images with different filtered disparities to generate a disparity image; including,
The synthesis filter applies a Gabor filter,
The kernel function setter includes a kernel function averaging unit that arithmetic averages kernel functions obtained by changing a directional parameter as a filter parameter defining a kernel function of the Gabor filter according to a rule mode or an irregular mode,
The disparity image generating apparatus using a synthesis filter, characterized in that when the kernel function averaging unit performs arithmetic average according to the irregular mode, the average kernel function (gavr-n) is obtained using the following equation.

Here, x and y are coordinate values in the rectangular coordinate system of the target image, λ is the wavelength that controls the sine function of the Gabor filter kernel, θ is the directionality of the kernel function, ψ is the phase difference, σ is the standard deviation of the Gaussian function, and γ is the Gabor It is a factor that determines the shape of the filter, and n+1 is the number of division for arithmetic mean.
delete delete delete delete delete delete delete illuminating the surface of the object using a plurality of lamps having different irradiation angles and scanning the surface of the object with an image sensor to obtain images with different parallax;
A synthesis filter receives the images obtained in the image acquisition step and performs filtering operations, but the synthesis kernel function applied to the synthesis filter is set by combining different kernel functions by a kernel function setter for at least one filter parameter that determines kernel characteristics and outputting a filtered stereo image;
calculating disparity information existing between one image and another image by matching the filtered stereo image in the synthesis filter;
Including; generating a disparity image using disparity information obtained when matching the stereo image;
When the synthesis filter applies the Gabor filtering technique, the synthesis kernel function of the synthesis filter is set by arithmetic average of kernel functions defined by the directionality parameter that determines the directionality of the Gabor filter,
When setting the arithmetic mean division number for dividing the combined kernel function when setting the synthetic kernel function, the same number as the number specified by the user is used according to the rule mode, or the number specified by the user is higher than the number specified by the user according to the irregular mode. A method of generating a disparity image using a synthesis filter, characterized in that a certain number of large numbers are used.
delete delete

Images