KR102282756B1 - Apparatus and method for gaussian filtering - Google Patents

Abstract

The Gaussian filtering apparatus generates a pixel array value by arranging the mask area data by referring to an image buffer generating mask area data including a value of a pixel located on a mask among the input image and an index array corresponding to the size of the mask, , including a Gaussian filter core that generates a Gaussian filter value by applying Gaussian filtering to a pixel array value.

Description

Gaussian filtering apparatus and method {APPARATUS AND METHOD FOR GAUSSIAN FILTERING}

The present invention relates to a filtering technique for an image, and more particularly, to a Gaussian filtering technique for an image.

An image recognition system that recognizes an object in an image, such as a face and gesture recognition system, is implemented based on software in various applications such as digital cameras, smart phones, access systems, and digital door locks as well as industrial systems. However, recently, as video input/output devices have been advanced, the number of systems supporting from a small QVGA level to a large HD level is increasing. The process of recognizing an object in an image includes Gaussian filtering.

That is, it is required to perform Gaussian filtering on a high-resolution image in real time, but since Gaussian filtering is performed depending on a processor through software in a general embedded system, there is a limit to real-time processing of Gaussian filtering on a high-resolution image. there is.

An object of the present invention is to provide a Gaussian filtering apparatus and method for performing Gaussian filtering through hardware.

According to an aspect of the present invention, there is provided an image buffer comprising: an image buffer for generating mask area data including pixel values of pixels located on a mask in an input image; and a Gaussian filter core configured to generate a pixel array value by arranging the mask region data with reference to an index array corresponding to the size of the mask, and to generate a Gaussian filter value by applying Gaussian filtering to the pixel array value A Gaussian filtering device is provided.

The image buffer may include: a plurality of serially connected first delay elements that sequentially receive and store pixel values of the input image and output the stored pixel values of the input image to the Gaussian filter core; and a plurality of image row storage units connected in series for outputting pixel values corresponding to the single row to the Gaussian filter core when all pixel values corresponding to a single row of the input image are received. At least one of the delay elements may output a pixel value of the input image to another first delay element, and the image row storage unit may output pixel values corresponding to the single row to another image row storage unit.

The image row storage unit may include: a plurality of serially connected second delay elements that sequentially receive and store pixel values corresponding to the single row and output the stored pixel values corresponding to the single row to the Gaussian filter core; and when all pixel values corresponding to the single row are received from the other image row storage unit or from the outside, the pixel values corresponding to the single row are sequentially stored in any one of the other image row storage unit, the second delay element, and a FIFO module sequentially outputting to the Gaussian filter core; At least one of the plurality of second delay elements may output the pixel value of the input image to another second delay element.

The Gaussian filter core includes: an index generator for generating a row index and a column index with reference to an index array corresponding to the size of the mask; a plurality of routers arranging some of the mask area data according to the row index and column index to generate pixel array values, respectively; and a filter core configured to generate a Gaussian filter value by applying Gaussian filtering to a matrix including each of the pixel array values in one row. may include.

The index generator may generate the row index and the column index including an index corresponding to each row corresponding to a mask center coordinate in the index array.

The int(n/2)+1th row of the index array is a row in which values 0 to n-1 are sequentially arranged, and each indices positioned in the int(n/2)+1th column of the index array are int With a value of (n/2), each index located above the int(n/2)+1th row among the indices of the index array has an index column of (int(n/2)+1-index row ), it has a value 1 larger than the index adjacent to the right, and when the index column is (int(n/2)+1-index row) or more, it has a value smaller than the index adjacent to the right by 1, the index Among the indices of the array, each index located below the int(n/2)+1th row is greater than the index adjacent to the left when the index column is greater than (int(n/2)+(n-index row)) If it has a value less than 1 and the index column is less than or equal to (int(n/2)+(n-index row)), has a value greater than the index adjacent to the left by 1, and n is the horizontal or vertical length of the mask , the index column may be a column number of each index, the index row may be a row number of each index, and the int(n/2) may be an integer obtained by n/2.

The router arranges pixel values corresponding to each index included in the column index among pixel values located in a row corresponding to the row index in the mask area data according to the position on the column index of each index to arrange the pixels value can be created.

According to another aspect of the present invention, in a method for a Gaussian filtering apparatus including an image buffer and a Gaussian filter core to perform Gaussian filtering on an input image, the image buffer is located on a mask of the input image. generating mask area data including values; generating, by the Gaussian filter core, a pixel array value by arranging the mask area data with reference to an index array corresponding to the size of the mask; and generating, by the Gaussian filter core, a Gaussian filter value by applying Gaussian filtering to the pixel array value.

The generating, by the image buffer, of the mask area data including the values of pixels located on the mask among the input images, may include sequentially receiving and storing pixel values of the input image, and storing the pixel values of the input image. a plurality of first delay elements connected in series for outputting to the Gaussian filter core; and a plurality of image row storage units connected in series for outputting pixel values corresponding to the single row to the Gaussian filter core when all pixel values corresponding to a single row of the input image are received. is performed, wherein at least one of the plurality of first delay elements outputs the pixel value of the input image to another first delay element, and at least one image row storage unit corresponds to the single row as another image row storage unit pixel values can be output.

The generating, by the image buffer, of the mask area data including the values of pixels located on the mask in the input image, may include sequentially receiving and storing pixel values corresponding to the single row, and storing the pixel values in the stored single row. a plurality of second delay elements connected in series to output corresponding pixel values to the Gaussian filter core; and when all pixel values corresponding to the single row are received from the other image row storage unit or from the outside, the pixel values corresponding to the single row are sequentially stored in one of the other image row storage unit, the second delay element, and a FIFO module sequentially outputting to the Gaussian filter core; The step is performed through the image buffer including an image row storage unit including a, and at least one of the plurality of second delay elements may output the pixel value of the input image to another second delay element.

The step of generating, by the Gaussian filter core, the pixel array value by arranging the mask area data with reference to the index array corresponding to the size of the mask, is performed by referring to the index array corresponding to the size of the mask through the index generator. creating indexes and column indexes; and generating pixel array values by arranging some of the mask area data according to the row index and column index through a plurality of routers; generating a Gaussian filter value by applying Gaussian filtering to a matrix including each of the pixel array values in one row through a filter core.

The generating of a row index and a column index by referring to the index array corresponding to the size of the mask through the index generator includes the row index including an index corresponding to each row corresponding to the mask center coordinate in the index array. and generating the column index.

The int(n/2)+1th row of the index array is a row in which values 0 to n-1 are sequentially arranged, and each indices positioned in the int(n/2)+1th column of the index array are int With a value of (n/2), each index located above the int(n/2)+1th row among the indices of the index array has an index column of (int(n/2)+1-index row ), it has a value 1 larger than the index adjacent to the right, and when the index column is (int(n/2)+1-index row) or more, it has a value smaller than the index adjacent to the right by 1, the index Among the indices of the array, each index located below the int(n/2)+1th row is greater than the index adjacent to the left when the index column is greater than (int(n/2)+(n-index row)) If it has a value less than 1 and the index column is less than or equal to (int(n/2)+(n-index row)), has a value greater than the index adjacent to the left by 1, and n is the horizontal or vertical length of the mask , the index column may be a column number of each index, the index row may be a row number of each index, and the int(n/2) may be an integer obtained by n/2.

The step of arranging some of the mask area data according to the row index and the column index through the plurality of routers to generate pixel array values may include: The step may be performed through the router for generating the pixel array value by arranging pixel values corresponding to each index included in the column index according to positions on the column index of each index.

As described above, according to an embodiment of the present invention, Gaussian filtering on a high-resolution image may be performed in real time.

1 is a block diagram illustrating a Gaussian filtering apparatus according to an embodiment of the present invention.
2 is a diagram illustrating a Gaussian distribution used for Gaussian filtering by a Gaussian filtering apparatus according to an embodiment of the present invention.
3 is a diagram illustrating a mask used by a Gaussian filtering apparatus according to an embodiment of the present invention.
4 is a diagram illustrating mirroring performed by a Gaussian filtering apparatus according to an embodiment of the present invention.
5 is a diagram illustrating an image buffer of a Gaussian filtering apparatus according to an embodiment of the present invention.
6 is a diagram illustrating an image row storage unit included in an image buffer of a Gaussian filtering apparatus according to an embodiment of the present invention.
7 is a diagram illustrating a Gaussian filter core of a Gaussian filtering apparatus according to an embodiment of the present invention.
8 is a diagram illustrating an index array stored in an index generator of a Gaussian filter core of a Gaussian filtering apparatus according to an embodiment of the present invention.
9 is a diagram conceptually illustrating an operation process of a router included in a Gaussian filter core of a Gaussian filtering apparatus according to an embodiment of the present invention.
10 is a diagram illustrating a result of a router included in a Gaussian filter core of a Gaussian filtering apparatus according to an embodiment of the present invention performing mirroring according to a column index and a row index;
11 is a diagram illustrating a result of a router included in a Gaussian filter core of a Gaussian filtering apparatus according to an embodiment of the present invention performing mirroring according to a column index and a row index according to each mask center coordinate.
12 is a flowchart illustrating a process in which a Gaussian filtering apparatus according to an embodiment of the present invention performs Gaussian filtering.

Since the present invention can have various changes and can have various embodiments, specific embodiments are illustrated in the drawings and will be described in detail through the detailed description. However, this is not intended to limit the present invention to specific embodiments, and it should be understood to include all modifications, equivalents and substitutes included in the spirit and scope of the present invention.

In addition, in this specification, when it is said that one component "transmits" a signal to another component, the one component may be directly connected to the other component to transmit a signal, but there is a particularly opposite description. Unless otherwise noted, it should be understood that a signal may be transmitted through another component in the middle.

1 is a block diagram illustrating a Gaussian filtering apparatus according to an embodiment of the present invention, FIG. 2 is a diagram illustrating a Gaussian distribution used by the Gaussian filtering apparatus according to an embodiment of the present invention for Gaussian filtering, and FIG. 3 is a diagram illustrating a mask used by the Gaussian filtering apparatus according to an embodiment of the present invention, and FIG. 4 is a diagram illustrating mirroring performed by the Gaussian filtering apparatus according to an embodiment of the present invention.

1 illustrates a Gaussian filtering apparatus according to an embodiment of the present invention including an image buffer 110 , a Gaussian filter core 120 , and a controller 130 .

The image buffer 110 receives input data that is an image received from an external device (a camera, a device connected through a network, a storage medium, etc.), and a value of a pixel corresponding to a mask for Gaussian filtering among the input data (hereinafter, a mask). area data) is transmitted to the Gaussian filter core 120 .

The Gaussian filter core 120 performs Gaussian filtering on the mask region data. In this case, when mirroring is required for the mask area data, the Gaussian filter core 120 may apply the mirroring to the mask area data to perform Gaussian filtering on the mask area data to which the mirroring is applied.

The Gaussian filter core 120 performs Gaussian filtering using a characteristic of a Gaussian distribution. The one-dimensional Gaussian distribution may be defined as in Equation 1 below.

[Equation 1]

는 각 원소값의 평균이고,

는 각 원소값의 표준편차이다.

인 경우, 1차원 가우시안 분포는 일반적인 정규분포가 될 수 있다. In this case, x means an element value (each pixel value of the input image) subject to Gaussian filtering,

is the average of each element value,

is the standard deviation of each element value.

In the case of , the one-dimensional Gaussian distribution may be a general normal distribution.

인 가우시안 분포에서 각 원소의 값은 범위

내에 분포되며, 표준 편차의 크기가 더 클수록, 더 큰 스무싱(smoothing) 현상이 일어난다(각 원소가 더 넓은 범위 내에 분포된다). In this case, the width of the Gaussian distribution may be determined by the standard deviation, and the size of the filter may be determined according to the width of the Gaussian distribution. 2, the mean is 0 and the standard deviation is

In a Gaussian distribution, the value of each element is a range

, and the larger the standard deviation, the greater the smoothing (each element is distributed within a wider range).

Since the Gaussian filtering apparatus according to an embodiment of the present invention performs Gaussian filtering on an image, a two-dimensional Gaussian distribution according to Equation 2 modified from Equation 1 may be used.

[Equation 2]

In this case, x is the x-coordinate of each element constituting the Gaussian distribution, and y is the y-coordinate of each element constituting the Gaussian distribution.

인 가우시안 분포에서 각 원소의 값은 범위

에 존재하기 때문에 마스크의 크기는 (8

+1) x (8

+1)로 설정될 수 있다. 따라서, 표준 편차가 1인 경우, 마스크는 9x9 크기의 마스크일 수 있다. 표준 편차가 0.4인 경우, 마스크는 3x3 크기의 마스크일 수 있다. 표준 편차가 0.8인 경우, 마스크는 5x5 크기의 마스크일 수 있다. Although the Gaussian function representing the Gaussian distribution is a continuous function in the Gaussian filtering apparatus according to an embodiment of the present invention, the value of the Gaussian function may be calculated only when x and y are integers in order to create a mask. mean is 0 and standard deviation is

In a Gaussian distribution, the value of each element is a range

The size of the mask is (8) because it exists in

+1) x (8

+1) can be set. Therefore, when the standard deviation is 1, the mask may be a 9x9 mask. If the standard deviation is 0.4, the mask may be a 3x3 mask. When the standard deviation is 0.8, the mask may be a 5x5 mask.

At this time, the Gaussian filtering process using a mask of a size determined according to the standard deviation calculates the sum of values obtained by multiplying the filter coefficients for each pixel of the input image corresponding to the mask at each position while moving the mask on the input image. It is a process.

When the position of the mask is as shown in (a) of FIG. 3 , all pixels of the input image corresponding to the mask exist. However, when the position of the mask is located as shown in FIG. 3B , there is a region where the pixel of the input image is not located. Therefore, for efficient Gaussian filtering processing, it is necessary to position the pixels of the input image with respect to all positions in the mask through mirroring as shown in FIG. 4 .

When mirroring is performed on the input image shown in (a) of FIG. 4 in the vertical direction, as shown in (b) of FIG. 4, rows corresponding to the specified range in the uppermost row of the input image are positioned at the top It may be copied to the uppermost row of the input image, and rows corresponding to a predetermined range in the lowermost row of the input image may be copied to the lowermost row of the input image. In addition, when mirroring is performed on the input image shown in (a) of FIG. 4 in the left and right directions, columns corresponding to the specified range in the leftmost column of the input image as shown in (c) of FIG. 4 are the leftmost columns. is copied to the left of the column located in , and columns corresponding to a range specified in the rightmost column of the input image as shown in FIG. 4C may be copied to the right of the rightmost column. In this case, the number of rows or columns to be copied may be (mask size-1)/2 or less.

As described above, the calculation complexity is high because mirroring requires calculating the copied position according to the position of each pixel. The Gaussian filter core 120 of the Gaussian filtering apparatus according to an embodiment of the present invention enables real-time Gaussian filtering by performing mirroring through hardware rather than a processor. A detailed structure of the Gaussian filter core 120 will be described in detail later with reference to FIGS. 7 to 9 .

The controller 130 transmits mask size information indicating the size of the mask to the Gaussian filter core 120 so that the Gaussian filter core 120 adjusts the mask area data output from the image buffer 110 to the size of the mask. control to extract. In this case, the controller 130 may receive an input for setting the mask size from the user and generate mask size information in advance.

5 is a diagram illustrating an image buffer of a Gaussian filtering apparatus according to an embodiment of the present invention, and FIG. 6 is a diagram illustrating an image row storage unit included in the image buffer of a Gaussian filtering apparatus according to an embodiment of the present invention am.

Referring to FIG. 5 , the image buffer 110 includes a plurality of delay elements 510 and a plurality of image row storage units 520 .

The delay element 510 stores a pixel value of a single pixel. When a new pixel value is received from the outside, each delay element 510 outputs a previously stored pixel value and stores the new pixel value. In this case, the plurality of delay elements 510 may be sequentially disposed with each other as shown in FIG. 5 . In this case, each delay element 510 may output a previously stored pixel value to a delay element (hereinafter referred to as a next delay element) corresponding to the following order and the Gaussian filter core 120 . In particular, since there is no next delay element for the last delay element among the plurality of delay elements 510 , the last delay element outputs the previously stored pixel value to the Gaussian filter core 120 .

The image row storage unit 520 stores a single row of the input image, and outputs a pixel value corresponding to an area in which a mask is located among pixel values included in the corresponding row. In this case, the plurality of image row storage units 520 may be sequentially arranged with each other. The image row storage unit 520 transmits the corresponding pixel values to the other adjacent image row storage unit 520 when other adjacent image row storage units exist and all pixel values corresponding to each row of the image are received. Accordingly, the image row storage 520 located at the top of the image row storage 520 illustrated in FIG. 5 stores the row of the input image input at the earliest time, and outputs the pixel values included in the row. can

Referring to FIG. 6 , the image buffer includes a FIFO module 610 and a plurality of delay elements 620 .

A first-in first-out (FIFO) module ( ) stores a pixel value corresponding to a single row received from an external or image row storage unit 520 . When the FIFO module 610 receives all pixel values corresponding to a single row, the delay element 620 connected to the FIFO module 610 and other image rows connected to the image row storage unit 520 store the corresponding pixel values. It transmits sequentially (first-in, first-out) to the FIFO module 610 of the unit 520 .

The plurality of delay elements 620 may be sequentially disposed with each other as shown in FIG. 5 . At this time, each delay element 620 may output the previously stored pixel value to the next delay element and the Gaussian filter core 120 . In particular, since there is no next delay element for the last delay element among the plurality of delay elements 620 , the last delay element outputs the previously stored pixel value to the Gaussian filter core 120 .

In this case, the Gaussian filter core 120 may perform Gaussian filtering by selecting mask region data that is a part of the outputs of the delay element 510 and the image row storage unit 520 according to the size of the mask. For example, when the size of the mask is 3x3, the Gaussian filter core 120 selects pixel values output from each image row storage unit 520 according to the path illustrated as 3x3 in FIG. 6 to perform Gaussian filtering. can In addition, when the size of the mask is 3x3, the Gaussian filter core 120 selects the pixel value output without passing through the delay element 510 and the pixel value output through the first two delay elements 510 as shown in FIG. 5 . Thus, Gaussian filtering can be performed. In addition, when the size of the mask is 3x3 as shown in FIG. 5 , the Gaussian filter core 120 includes a pixel value output without passing through the delay element 510 , a pixel value output through the delay element 510 , and two lower images. Some of the pixel values output through the row storage unit 520 may be selected according to the above-described process.

The Gaussian filter core 120 adaptively performs Gaussian filtering with respect to the mask size by receiving only pixel values output through the paths as illustrated in FIGS. 5 and 6 even when the size of the mask is 5x5, 7x7, or 9x9. can be done In this case, the controller 130 may transmit a mask size setting signal for setting the size of the mask to the Gaussian filter core 120 , and the Gaussian filter core 120 may select a pixel value according to the mask size setting signal. .

Accordingly, a pixel value received by the Gaussian filter core 120 among pixel values output from the image buffer 110 corresponds to the above-described mask region data. The image buffer 110 periodically outputs a different path through which the pixel value of the input image is output through the above-described structure, and the Gaussian filter core 120 selects the pixel value according to the path through which each pixel value is output. A pixel value of the input image corresponding to the position of each mask may be obtained.

7 is a diagram illustrating a Gaussian filter core of a Gaussian filtering apparatus according to an embodiment of the present invention, and FIG. 8 is an index array stored in an index generator of a Gaussian filter core of a Gaussian filtering apparatus according to an embodiment of the present invention. 9 is a diagram conceptually illustrating an operation process of a router included in a Gaussian filter core of a Gaussian filtering apparatus according to an embodiment of the present invention, and FIG. 10 is a Gaussian according to an embodiment of the present invention. It is a diagram illustrating a result of a router included in a Gaussian filter core of a filtering device performing mirroring according to a column index and a row index. 11 is a diagram illustrating a result of a router included in a Gaussian filter core of a Gaussian filtering apparatus according to an embodiment of the present invention performing mirroring according to a column index and a row index according to each mask center coordinate.

Referring to FIG. 7 , the Gaussian filter core 120 of the Gaussian filtering apparatus includes an index generator 710 , a plurality of routers 720 , and a filter core 730 .

The index generator 710 receives mask size information from the controller 130 and generates an index, which is a reference for rearranging pixel values received from the image buffer 110 by each router 720 according to the size and position of the mask. to each router 720 . In this case, the index generator 710 may transmit an index corresponding to the position of the mask to each router 720 with reference to the index array stored in advance. Referring to FIG. 8 , the index array may be an array including an index set according to the size of the mask (n x n, where n is an odd number equal to or greater than 3). The int(n/2)+1-th row of the index array (hereinafter, referred to as a steady-state row) may be a row in which values 0 to n-1 are sequentially arranged. In addition, each index (hereinafter, referred to as a middle column index) positioned in the int(n/2)+1th column of the index array may have a value of int(n/2). In addition, among the indices of the index array, the indices located above the steady state row may have a value greater than the index adjacent to the right by 1 when the index column is smaller than (int(n/2)+1-index row), , if the index column is equal to or greater than (int(n/2)+1-index row), it may have a value smaller by 1 than the index adjacent to the right. In addition, among the indices of the index array, the indices located at the lower side of the steady-state row have a value smaller than the index adjacent to the left by 1 when the index column is larger than (int(n/2)+(n-index row)). and, when the index column is less than or equal to (int(n/2)+(n-index row)), it may have a value 1 greater than the index adjacent to the left. In this case, the index column indicates the column number of each index, and the index row indicates the row number of each index. Also, int(a) means a function that converts a real number a into an integer.

The index generator 710 is an index array corresponding to the coordinates (hereinafter, referred to as mask center coordinates) at which the center point of the mask (a point corresponding to the coordinate int(n/2)+1 on the mask) is located on the image of the input data. A row index and a column index corresponding to a row of are extracted and transmitted to the router 720 . For example, when the mask center coordinates are (a, b), the index generator 710 may determine whether each a or b is equal to or less than int(n/2). If a is less than or equal to int(n/2), the index generator 710 may transmit an index corresponding to the a-th row of the index array to the router 720 as a row index. If a is greater than int(n/2) and a is equal to or less than (h-int(n/2), h is the height of the image), the index generator 710 sets the index (0 to 0 to) corresponding to the steady-state row. n-1) may be transmitted to the router 720 . If a is greater than h-int(n/2), the index generator 710 may output an index corresponding to the (n-a)-th row of the index array as a row index. If b is less than or equal to int(n/2), the index generator 710 may transmit the index corresponding to the b-th row of the index array to the router 720 as a column index. If b is greater than int(n/2) and b is less than or equal to (w-int(n/2), w is the image area), the index generator 710 generates indices (0 to 0 to) corresponding to the steady-state row. n-1) may be transmitted to the router 720 as a column index. If b is greater than w-int(n/2), the index generator 710 may output an index corresponding to the (n-a)-th row of the index array as a column index.

The router 720 arranges and outputs the mask area data that is the pixel value received from the image buffer 110 based on the row index and column index received from the index generator 710 . Referring to FIG. 9( a ), the router 720 receives a plurality of pixel values (81 pixel values, inData_0 to inData_80 when the size of the mask is 9x9) from the image buffer 110, and the index generator 710 A pixel array value that is a value obtained by arranging one or more of the pixel values input from the image buffer 110 is generated according to the row index and column index received from , and the pixel array value is transmitted to the filter core 730 . In this case, the router 720 may be configured in plurality, and each router 720 receives all pixel values, and then generates a pixel array value in which some of the total pixel values are arranged according to the row index and column index. to the filter core 730 , respectively. For example, when each router 720 receives row index 0 and column index {8, 7, 6, 5, 4, 5, 6, 7, 8} as shown in FIG. 9(b), row index 0 A pixel array value may be generated by arranging some of pixel values (pixel values corresponding to column index 4 to column index 8) corresponding to the row corresponding to the column index to correspond to the column index. That is, the router 720 determines that a pixel value corresponding to row index 0 (corresponding to the first row among all pixel values) among the received pixel values is (a, b, c, d, e, f, g, h, i). ), which is (i, h, g, f, e, f, g, h, i) can be created as a pixel array value. As described above, each router 720 may output one row of pixel array values to which mirroring is applied. Accordingly, when a 3×3 mask is applied, three routers 720 may generate a pixel array value, and when a 9×9 mask is applied, nine routers 720 may generate a pixel array value. Referring to FIG. 10 , when the mask center coordinate is (1,1), each router 720 may receive any one of a column index 1002 and a row index 1003 to generate a pixel array value. Accordingly, if the pixel array values output by each router 720 are represented as a matrix as shown in FIG. 10, the mirroring may be performed. That is, even if the position of the mask is changed so that the center point of the mask, which is a pixel indicated by a circle, is changed as shown in FIG. 11 , each router 720 can adaptively perform mirroring for Gaussian filtering. At this time, the mask center coordinates for FIG. 12(a) are (1, 1), the mask center coordinates for FIG. 12(b) are (1,2), and the mask center coordinates for FIG. 12(c) are (1, 3).

The filter core 730 generates a Gaussian filter value by applying Gaussian filtering to a matrix including each pixel array value in one row. In this case, the filter core 730 may generate a Gaussian filter value according to a known Gaussian filtering method.

12 is a flowchart illustrating a process in which the Gaussian filtering apparatus performs Gaussian filtering according to an embodiment of the present invention.

Referring to FIG. 12 , in operation 1210 , the image buffer 110 receives input data.

In operation 1220 , the image buffer 110 transmits mask region data among input data to the Gaussian filter core 120 . For example, the image buffer 110 transmits data of a pixel located on a mask among input data to the filter core 730 . A process in which the image buffer 110 outputs mask area data has been described above with reference to FIGS. 5 and 6 .

In operation 1230, the Gaussian filter core 120 generates a row index and a column index according to the mask center coordinates. For example, the Gaussian filter core 120 may set indices corresponding to rows corresponding to the x-coordinates and y-coordinates of the mask center coordinates among the preset index arrays as row indexes and column indices, respectively. The process of generating the row index and the column index by the Causian filter core 120 has been described above with reference to FIGS. 7 to 8 .

In operation 1240, the Gaussian filter core 120 generates a pixel array value by arranging pixel values of the mask area data according to a row index and a column index. A detailed process for generating the pixel array value by the Gaussian filter core 120 has been described above with reference to FIGS. 9 to 11 .

In operation 1250, the Gaussian filter core 120 generates and outputs a Gaussian filter value by applying Gaussian filtering to the pixel array value.

Therefore, the Gaussian filtering apparatus and method according to an embodiment of the present invention perform the Gaussian filtering process and the mirroring process, which is a pre-processing of the Gaussian filtering, through a hardware structure, thereby reducing the time required for the Gaussian filtering through software. can be reduced compared to

So far, the present invention has been looked at with respect to the embodiment. Many embodiments other than the above-described embodiments are within the scope of the claims of the present invention. Those of ordinary skill in the art to which the present invention pertains will understand that the present invention can be implemented in a modified form without departing from the essential characteristics of the present invention. Therefore, the disclosed embodiments should be considered in an illustrative rather than a restrictive sense. The scope of the present invention is indicated in the claims rather than the foregoing description, and all differences within the scope equivalent thereto should be construed as being included in the present invention.

Claims (14)

an image buffer for generating mask area data including pixel values of pixels positioned on a mask in an input image; and
A Gaussian filter core that generates a pixel array value by arranging the mask area data with reference to an index array corresponding to the size of the mask, and applies Gaussian filtering to the pixel array value to generate a Gaussian filter value,
The Gaussian filter core comprises:
an index generator for generating a row index and a column index by referring to an index array corresponding to the size of the mask;
a plurality of routers arranging some of the mask area data according to the row index and column index to generate pixel array values, respectively; and
and a filter core configured to generate a Gaussian filter value by applying Gaussian filtering to a matrix including each of the pixel array values in one row.
According to claim 1,
the video buffer
a plurality of serially connected first delay elements that sequentially receive and store pixel values of the input image and output the stored pixel values of the input image to the Gaussian filter core; and
A plurality of serially connected image row storage unit for outputting pixel values corresponding to the single row to the Gaussian filter core when all pixel values corresponding to a single row of the input image are received
including,
at least one of the plurality of first delay elements outputs a pixel value of the input image to another first delay element,
The image row storage unit outputs one or more pixel values corresponding to the single row to another image row storage unit.
3. The method of claim 2,
The image row storage unit,
a plurality of serially connected second delay elements that sequentially receive and store pixel values corresponding to the single row and output the stored pixel values corresponding to the single row to the Gaussian filter core; and
When all pixel values corresponding to the single row are received from the other image row storage unit or the outside, the pixel values corresponding to the single row are sequentially stored in the other image row storage unit, any one of the second delay element, and the a FIFO module that sequentially outputs to a Gaussian filter core;
including,
At least one of the plurality of second delay elements outputs a pixel value of the input image to another second delay element.
delete According to claim 1,
The index generator generates the row index and the column index including an index corresponding to each row corresponding to the mask center coordinates in the index array.
6. The method of claim 5,
The int(n/2)+1th row of the index array is a row in which values of 0 to n-1 are sequentially arranged,
Each index located in the int(n/2)+1th column of the index array has a value of int(n/2),
Among the indices of the index array, each index located above the int(n/2)+1-th row is smaller than the index adjacent to the right when the index column is smaller than (int(n/2)+1-index row). If it has a value greater than 1 and the index column is (int(n/2)+1-index row) or more, it has a value less than the index adjacent to the right by 1,
Among the indices of the index array, each index located below the int(n/2)+1th row is adjacent to the left when the index column is greater than (int(n/2)+(n-index row)) If it has a value less than the index by 1, and the index column is less than or equal to (int(n/2)+(n-index row)), it has a value 1 greater than the index adjacent to the left,
Wherein n is the horizontal or vertical length of the mask, the index column is the column number of each index, the index row is the row number of each index, and the int(n/2) is an integer number of n/2. Gaussian filtering device, characterized in that.
According to claim 1,
The router arranges pixel values corresponding to each index included in the column index among pixel values located in a row corresponding to the row index in the mask area data according to the position on the column index of each index to arrange the pixels Gaussian filtering device, characterized in that for generating a value.
A method for performing Gaussian filtering on an input image by a Gaussian filtering apparatus including an image buffer and a Gaussian filter core, the method comprising:
generating, by the image buffer, mask area data including pixel values of pixels located on a mask in the input image;
generating, by the Gaussian filter core, a pixel array value by arranging the mask area data with reference to an index array corresponding to the size of the mask; and
and generating, by the Gaussian filter core, a Gaussian filter value by applying Gaussian filtering to the pixel array value,
generating, by the Gaussian filter core, a pixel array value by arranging the mask area data with reference to an index array corresponding to the size of the mask,
generating a row index and a column index by referring to an index array corresponding to the size of the mask through an index generator;
generating pixel array values by arranging some of the mask area data according to the row index and column index through a plurality of routers; and
and generating a Gaussian filter value by applying Gaussian filtering to a matrix including each of the pixel array values in one row through a filter core.
9. The method of claim 8,
generating, by the image buffer, mask area data including a value of a pixel located on a mask among the input image,
a plurality of serially connected first delay elements that sequentially receive and store pixel values of the input image and output the stored pixel values of the input image to the Gaussian filter core; and
A plurality of serially connected image row storage unit for outputting pixel values corresponding to the single row to the Gaussian filter core when all pixel values corresponding to a single row of the input image are received
It is a step performed through the image buffer comprising a,
at least one of the plurality of first delay elements outputs a pixel value of the input image to another first delay element,
The image row storage unit outputs the pixel values corresponding to the single row to one or more other image row storage units.
10. The method of claim 9,
generating, by the image buffer, mask area data including a value of a pixel located on a mask among the input image,
a plurality of serially connected second delay elements that sequentially receive and store pixel values corresponding to the single row and output the stored pixel values corresponding to the single row to the Gaussian filter core; and
When all pixel values corresponding to the single row are received from the other image row storage unit or the outside, the pixel values corresponding to the single row are sequentially stored in the other image row storage unit, any one of the second delay element, and the a FIFO module that sequentially outputs to a Gaussian filter core;
It is a step performed through the image buffer comprising an image row storage comprising a,
At least one of the plurality of second delay elements outputs a pixel value of the input image to another second delay element.
delete 9. The method of claim 8,
The step of generating a row index and a column index by referring to an index array corresponding to the size of the mask through the index generator,
and generating the row index and the column index including an index corresponding to each row corresponding to a mask center coordinate in the index array.
13. The method of claim 12,
The int(n/2)+1th row of the index array is a row in which values of 0 to n-1 are sequentially arranged,
Each index located in the int(n/2)+1th column of the index array has a value of int(n/2),
Among the indices of the index array, each index located above the int(n/2)+1-th row is smaller than the index adjacent to the right when the index column is smaller than (int(n/2)+1-index row). If it has a value greater than 1 and the index column is (int(n/2)+1-index row) or more, it has a value less than the index adjacent to the right by 1,
Among the indices of the index array, each index located below the int(n/2)+1th row is adjacent to the left when the index column is greater than (int(n/2)+(n-index row)) If it has a value less than the index by 1 and the index column is less than or equal to (int(n/2)+(n-index row)), it has a value 1 greater than the index adjacent to the left,
Wherein n is the horizontal or vertical length of the mask, the index column is a column number of each index, the index row is a row number of each index, and the int(n/2) is an integer number of n/2. Gaussian filtering method, characterized in that.
9. The method of claim 8,
generating pixel array values by arranging some of the mask area data according to the row index and column index through the plurality of routers,
The pixel array value is generated by arranging pixel values corresponding to each index included in the column index among pixel values positioned in a row corresponding to the row index in the mask area data according to the position on the column index of each index. Gaussian filtering method, characterized in that the step performed through the router.

Images