TWI479448B - Image pre-processing system and method - Google Patents

Description

Image pre-processing method and system

The present invention relates to an image pre-processing method and system, and more particularly to an image pre-processing method and system for improving efficiency by matrix pre-processing.

With the increasing popularity of digital images, image processing and analysis is also widely used in the fields of medicine, mathematics, biology, meteorology, engineering science, etc., and even in daily life. Recently, due to the advancement of multimedia device technology, image processing speed has increased dramatically and processing costs have decreased, and display image processing technology has become increasingly important.

A digital image is a collection of points (ie, pixels) having different brightness and color. Each pixel has its position and gray level by using the position and gray value of each pixel in the digital image. In the form of a matrix, different algorithms can be used to achieve various image processing functions, such as image conversion, color conversion and analysis, image enhancement, feature extraction, image segmentation, image representation and description, image compression and Image reconstruction, etc.

However, since the conventional image processing method needs to use the fourth-order array operation, it is not conducive to the high-efficiency second-order matrix operation of a large number of core processors in the central processing unit, the multi-core central processing unit and the display chip, so that the image processing efficiency is The image size has risen sharply and has fallen sharply. As shown in the following fourth-order array equation (1) of image processing, where H _ij represents a matrix after image processing, F _ijkl represents a fourth-order matrix of a region to be processed in the original image, and G _kl represents an image to be set. A mask matrix formed by the processing algorithm, m and n are positive integers for indicating the size of the mask matrix.

It can be seen that if the central processor performs the operation on the above equation (1), the efficiency of image processing cannot be improved.

In summary, how to improve the array operation in image processing to improve the efficiency of image processing has become an urgent problem to be solved.

In order to solve the above-mentioned shortcomings of the prior art, an object of the present invention is to provide an image pre-processing method and system, which can utilize the high-performance second-order matrix computing capability of the display chip by converting the fourth-order array of image processing into a second-order array. Can greatly improve the efficiency of image processing.

The invention discloses an image pre-processing method, comprising the steps of: (A) obtaining an original image matrix having a size of [MN]; and (B) converting a preset image processing algorithm into a first mask having a size of [mn] a mask matrix; (C) taking a position of any element in the original image matrix as an origin, selecting a matrix of the area to be tested of [mn] size, and then adding a blank array around the original image matrix to form mxn Expanding the matrix; (D) forming a temporary storage matrix having a size of [MxN mxn] according to the size of the first mask matrix and the size of the original image matrix, wherein the element MxN is used to memorize the original matrix of the area to be tested Point, element mxn is used to memorize the elements of each matrix of the area to be tested; (E) repeat steps (C) and (D) Until all elements in the original image matrix are calculated to fill the temporary storage matrix; (F) expanding the first mask matrix corresponding to the temporary storage matrix to a second mask matrix having a size of [MxN mxn]; (G) multiplying the temporary storage matrix by the second mask matrix to obtain a result matrix of [MxN mxn] size, and then performing the step conversion according to a predetermined rule to obtain a new image matrix.

In the foregoing image preprocessing method, the image processing calculation formula may be an image filtering calculation formula, a morphological calculation formula or a corner detection calculation formula, and the parallel operation technology may be used to perform the temporary storage matrix and the second mask matrix. Point multiply.

The present invention further discloses an image pre-processing method comprising the steps of: (A) obtaining an original image matrix having a size of [MN]; (B) using the original image matrix as an initial matrix, further obtaining the original image matrix to change with time. Complex image matrix to be observed; (C) selecting a feature region in the original image matrix, and converting the feature region into a first mask matrix having a size of [mn]; (D) using any of the original image matrix The position of the element is the origin, a matrix of the area to be tested of [mn] size is selected, and a blank array is added around the original image matrix to form mxn expansion matrices; (E) according to the size of the first mask matrix and The size of the original image matrix forms a temporary storage matrix having a size of [MxN mxn], wherein the element MxN is used to memorize the position of the origin selected by the original image matrix, and the element mxn is used to memorize each of the regions to be tested. The elements of the matrix; (F) repeating steps (D) and (E) until all elements in the original image matrix are calculated to fill the temporary storage matrix; (G) the first mask matrix corresponds to the temporary Deposit moment Expanded to a second mask matrix having a size of [MxN mxn]; (H) multiply the scratch matrix with the second mask matrix to obtain a result matrix of [MxN mxn] size, and then the result matrix Performing gradation conversion according to a predetermined rule to obtain a position of the feature region in the original image; and (G) replacing the original image matrix with each of the image matrix to be observed, performing step (D) to step (H) to obtain The feature area is located in each of the image matrix to be observed, and a trajectory of the feature area moving with time is established.

The image pre-processing system of the present invention comprises: an original image module for acquiring an original image matrix having a size of [MN]; and an image processing algorithm module for converting a preset image processing algorithm into having [ Mn] the first mask matrix of size; the expansion matrix module takes the position of any element in the original image matrix as the origin, selects a matrix of the area to be tested of [mn] size, and surrounds the original image matrix The blank array is added to form mxn expansion matrices; the temporary storage matrix module forms a temporary storage matrix having a size of [MxN mxn] according to the size of the first mask matrix and the size of the original image matrix, wherein the element MxN The system is used to memorize the origin of the matrix of the area to be tested, and the element mxn is used to memorize the elements of the matrix of the area to be tested; the loop processing module is used to repeatedly execute the expansion matrix module and the temporary storage matrix module. The function of the group until all the elements in the original image matrix are calculated to fill the temporary storage matrix; the mask matrix processing module is used to expand the first mask matrix corresponding to the temporary storage matrix to have [ a second mask matrix of size MXN mxn; and an image processing module for dot-multiplying the temporary storage matrix and the second mask matrix to obtain a result matrix of [MxN mxn] size, and then The result matrix is gradually converted according to a predetermined rule to obtain a new image. Like a matrix.

Compared with the prior art, the image processing method and system of the present invention performs matrix pre-processing by reducing the loop required for matrix multiplication in image processing, and the parallel computing technique can significantly improve the efficiency of image processing.

The embodiments of the present invention are described below by way of specific embodiments, and those skilled in the art can readily appreciate the other advantages and advantages of the present invention. The invention may also be embodied or applied by other different embodiments.

1 is a flow chart of the image pre-processing method of the present invention. As shown in the figure, the specific process includes the following steps: In step S10, a second-order original image matrix having a size of [MN] is obtained, which can be represented by A _ij Where, i =1...M and j =1...N, M, N are positive integers.

In step S11, the pre-set image processing algorithm is converted into a second-order first mask matrix having a size of [mn], which can be represented by G _kl , where k =1...m and l =1...n, m, n Is a positive integer.

In a preferred embodiment, the image processing algorithm can be an image filtering algorithm, a morphological algorithm, or a corner detection algorithm.

In another embodiment, the image processing algorithm is applied to parallel operations of parallel computing and display wafers of a cluster computer.

In step S12, using a position of any element in the original image matrix as an origin, a matrix of the area to be tested of [mn] size is selected, which may be represented by F _ij , and a blank array is added around the original image matrix to form mxn. Expansion matrix. The detailed implementation process of step S12 is: first, a blank array of size [M(n-1)] is attached below the original image matrix, and the size is [(m-1)(N+(n-1)). a blank array is attached to the left of the original image matrix to obtain a first expansion matrix, and then a row of blank arrays is moved at a time to sequentially add the blank array to the left of the original image matrix to the right of the original image matrix Square to obtain the second to mth expansion matrix, and finally, move a column of blank arrays at a time to sequentially add the blank array below the original image matrix to the top of the original image matrix, and move a column of blank arrays at a time. Thereafter, the steps of obtaining the second to mth expansion matrices are repeatedly performed to obtain the (m+1)th to (mxn)th expansion matrices. Expansion matrix can It is expressed that its relationship with the matrix of the area to be tested is as shown in the following equation (2):

In step S13, a temporary storage matrix having a size of [MxN mxn] is formed according to the size of the first mask matrix and the size of the original image matrix, wherein the element MxN is used to store the selected image matrix. The position of the origin, the element mxn is used to memorize the elements of the matrix of the area to be tested in the expansion matrix.

In step S14, step S12 and step S13 are repeatedly performed until all the elements in the original image matrix are calculated to fill the temporary storage matrix, wherein the temporary storage matrix is It is indicated that the relationship between the temporary storage matrix and the expansion matrix is as shown in the following equation (3):

In step S15, the first mask matrix corresponding to the temporary storage matrix is expanded into a second mask matrix having a size of [MxN mxn], and the second mask matrix is Said.

In step S16, the temporary storage matrix is dot-multiplied with the second mask matrix, as shown in the following equation (4), to obtain a result matrix of [MxN mxn] size, and the result matrix is Representing, the result matrix is gradually converted according to a predetermined rule to obtain a new image matrix.

In a preferred embodiment, the step S16 can perform point multiplication of the temporary storage matrix and the second mask matrix by using a parallel computing technique, and the predetermined rule can be Equation (5) and Equation (6): p = ( k -1) n + l (5)

q =( i -1) n + j (6)

In another embodiment, the image pre-processing method of the present invention can be applied to a regional mean filtering algorithm or a regional median filtering algorithm to convert a fourth-order matrix operation into a second-order matrix operation, such as the following equations. (7.1) and (7.2) and equations (8.1) and (8.2):

H ( i , j )= median { F ( i + k , j + l );( p , q )} (8.1)

2a to 2f are schematic diagrams of an image pre-processing method to which the present invention is applied, and the specific flow includes the following: as shown in FIG. 2a, a second-order original image matrix 20 having a size of [MN] is first obtained, where M=512, N=512, [MN] has 512x512 elements, and the following definitions and representations for the matrix [MN] or [mn].

As shown in FIG. 2b, the pre-set image processing equation is converted into a [3 3] size (ie, the matrix [3 3] has 3x3=9 elements) and has a second-order first mask matrix 21 of elements G1 to G9. .

As shown in FIG. 2c, an element of the area of i=1, j=1 in the original image matrix 20 is used as an origin, and a matrix 22 of the area to be tested of [3 3] size is selected, wherein elements 1 to 9 are included. Then proceed with the following steps:

(1) First, a blank array of size [512(3-1)] is attached below the original image matrix, and a blank array of size [(3-1)(512+(3-1))]] Attached to the left of the original image matrix to obtain the first expansion matrix 221.

(2) Moving a row of blank arrays at a time to sequentially add the blank array to the left of the original image matrix to the right of the original image matrix to obtain the second to third expansion matrices 222 to 223.

(3) moving a column of blank arrays at a time to sequentially add the blank array below the original image matrix to the top of the original image matrix, and repeating step (2) after each moving of a column of blank arrays to obtain Fourth to ninth expansion matrices (224-229).

As shown in FIG. 2d, according to the size of the first mask matrix 21 and the size of the original image matrix 20, a temporary storage matrix 23 having a size of [512x512 3x3] is formed, wherein 512x512 elements in the p direction are formed. For storing the position of the origin selected in the original image matrix 20, 3x3 elements located in the q direction are used to memorize the elements 1 to 9 of the matrix of the area to be tested in the expansion matrices 221 to 229.

Then, the elements of the other positions in the original image matrix 20 are used as the origin, and the steps of the second and second graphs are repeatedly executed until all the elements in the original image matrix 20 are calculated to fill the temporary memory matrix 23, In order to make the original image region to be tested in the fourth-order array form in the conventional image processing method, the matrix can be converted into the second-order temporary storage matrix 23.

As shown in FIG. 2e, the first mask matrix 21 corresponds to the temporary memory matrix 23 and is expanded into a second mask matrix 24 having a size of [512x512 9x9].

As shown in FIG. 2f, the temporary storage matrix 23 and the second mask matrix 24 are dot-multiplied to obtain a result matrix 25 of a size of [512×512 9×9], and the result matrix is as in the above equations (5) and ( 6) The predetermined rule is gradually converted to obtain a new image matrix 26.

FIG. 3 is a flow chart of another embodiment of the image pre-processing method of the present invention, which can be applied to Digital Image Correlation (DIC). As shown in the figure, the specific process includes the following steps: In step S30, an original image matrix having a size of [M N] is obtained.

In step S31, the original image matrix is used as an initial matrix. The image matrix to be observed of the original image matrix changes over time.

In step S32, a feature region is selected in the original image matrix, and the feature region is converted into a first mask matrix having a size of [m n].

In step S33, a matrix of the area to be tested of [m n] size is selected with the position of any element in the original image matrix as an origin, and a blank array is added around the original image matrix to form mxn expansion matrices. The detailed implementation process of step S33 is: first, a blank array of size [M(n-1)] is attached below the original image matrix, and the size is [(m-1)(N+(n-1)). a blank array is attached to the left of the original image matrix to obtain a first expansion matrix, and then a row of blank arrays is moved at a time to sequentially add the blank array to the left of the original image matrix to the right of the original image matrix Square to obtain the second to mth expansion matrix, and finally, move a column of blank arrays at a time to sequentially add the blank array below the original image matrix to the top of the original image matrix, and move a column of blank arrays at a time. Thereafter, the steps of obtaining the second to mth expansion matrices are repeatedly performed to obtain the (m+1)th to (mxn)th expansion matrices.

In step S34, a temporary storage matrix having a size of [MxN mxn] is formed according to the size of the first mask matrix and the size of the original image matrix, wherein the element MxN is used to store the selected image matrix. The position of the origin, the element mxn is used to memorize the elements of the matrix of the area to be tested.

In step S35, step S33 and step S34 are repeatedly performed until all the elements in the original image matrix are calculated to fill the temporary storage matrix.

In step S36, the first mask matrix corresponding to the temporary storage matrix is expanded into a second mask matrix having a size of [MxN mxn].

In step S37, the temporary storage matrix is dot-multiplied with the second mask matrix to obtain a result matrix of [MxN mxn] size, and then the result matrix is subjected to gradation conversion according to a predetermined rule to obtain the feature region. The location in the original image.

Finally, in step S38, the original image matrix is replaced by each of the to-be-observed image matrices, and steps S32 to S37 are performed to obtain the position of the feature region in each of the to-be-observed image matrices, and the feature region is established to move with time. The trajectory.

Figure 4 is a graph comparing the efficiency of the image pre-processing method and the conventional image processing to which the present invention is applied. As shown in the figure, the horizontal axis is the size of the original image matrix, and the image processing is performed by the CPU and the GPU respectively under the condition that the size of the first mask matrix is [7 7], and the result shows the image pre-processing method of the present invention. It can effectively improve the efficiency of image processing, especially with the GPU's parallel computing ability. When the original image matrix size is [1024 1024], it can achieve about 12 times higher efficiency than the conventional image processing method.

Referring to FIG. 5, the present invention provides an image pre-processing system 5, including an original image module 50, an image processing algorithm module 51, an expansion matrix module 52, a temporary storage matrix module 53, and a loop processing module. 54. The mask matrix processing module 55 and the image processing module 56.

The original image module 50 is used to obtain an original image matrix having a size of [M N].

Image processing algorithm module 51 for processing pre-set image The calculus is converted to a first mask matrix of size [m n].

In a preferred embodiment, the image processing algorithm can be image filtering, morphology, or corner detection.

The expansion matrix module 52 selects a matrix of the area to be tested of [mn] size with the position of any element in the original image matrix as an origin, and then adds a blank array around the original image matrix to form mxn expansion matrices. .

The temporary storage matrix module 53 forms a temporary storage matrix having a size of [MxN mxn] according to the size of the first mask matrix and the size of the original image matrix, wherein the element MxN is used to memorize the original matrix. The position of the selected origin, the element mxn is used to memorize the elements of the matrix of the area to be tested.

The loop processing module 54 is configured to repeatedly execute the functions of the expansion matrix module and the temporary storage matrix module until all elements in the original image matrix are calculated to fill the temporary storage matrix.

The mask matrix processing module 55 is configured to expand the first mask matrix corresponding to the temporary storage matrix into a second mask matrix having a size of [MxN mxn].

The image processing module 56 is configured to multiply the temporary storage matrix and the second mask matrix to obtain a result matrix of [MxN mxn] size, and then perform the step conversion according to a predetermined rule to obtain the result matrix. New image matrix.

In summary, the image pre-processing method and system of the present invention reduces the fourth-order array operation in the conventional image processing technology to a second-order array operation, thereby avoiding the inability to effectively utilize the central processing unit or the display chip in the prior art. A large number of core processor issues to significantly improve the efficiency of image processing.

The above embodiments are merely illustrative of the principles and functions of the present invention. It is not intended to limit the invention. Modifications and variations of the above-described embodiments can be made by those skilled in the art without departing from the spirit and scope of the invention.

20‧‧‧ original image matrix

21‧‧‧First mask matrix

221~229‧‧‧First to Ninth Expansion Matrix

23‧‧‧Scratch matrix

24‧‧‧second mask matrix

25‧‧‧Result matrix

26‧‧‧New Image Matrix

5‧‧‧Image pre-processing system

50‧‧‧ original image module

51‧‧‧Image Processing Algorithm Module

52‧‧‧Expansion matrix module

53‧‧‧Scratch Matrix Module

54‧‧‧Circle processing module

55‧‧‧Mask matrix processing module

56‧‧‧Image Processing Module

S10~S16‧‧‧Steps

S30~S38‧‧‧Steps

1 is a flow chart of a pre-image processing method of the present invention; 2a to 2f are schematic views of an image pre-processing method to which the present invention is applied; and FIG. 3 is another embodiment of an image pre-processing method to which the present invention is applied. FIG. 4 is a comparison diagram of the efficiency of the image pre-processing method and the conventional image processing to which the present invention is applied; and FIG. 5 is an architectural diagram of the image pre-processing system of the present invention.

S10~S16‧‧‧Steps

Claims (7)

An image preprocessing method includes: (A) obtaining an original image matrix having a size of [MN]; (B) converting a preset image processing formula into a first mask matrix having a size of [mn]; Taking a position of any element in the original image matrix as an origin, selecting a matrix of the area to be tested of [mn] size, and then adding a blank array around the original image matrix to form mxn expansion matrices; (D) according to the The size of the first mask matrix and the size of the original image matrix form a temporary storage matrix having a size of [MxN mxn], wherein the element MxN is used to memorize the position of the origin selected by the original image matrix, the element mxn And (E) repeating steps (C) and (D) until all elements in the original image matrix are calculated to fill the temporary storage matrix; (F) The first mask matrix is correspondingly expanded into a second mask matrix having a size of [MxN mxn]; and (G) multiplying the temporary matrix by the second mask matrix to obtain [MxN Mxn] the result matrix of the size, and then the result matrix is scheduled Gradation conversion is performed to obtain a new image matrix. The image preprocessing method according to claim 1, wherein the image processing algorithm is an image filtering algorithm, a morphological calculation formula, a digital image correlation method, or a corner detection calculation formula. The image pre-processing method according to claim 1, wherein The image processing algorithm is applied to the parallel operation of parallel computing and display chips of a cluster computer. The image pre-processing method according to claim 1, wherein the step (C) comprises: (1) adding a blank array of [M(n-1)] size below the original image matrix, and Adding a blank array of [(m-1)(N+(n-1))] size to the left of the original image matrix to generate a first expansion matrix; (2) moving a row of blank arrays at a time to the original image matrix The left blank array is sequentially attached to the right of the original image matrix to obtain the second to mth expansion matrix; and (3) one column of the blank array is moved at a time to sequentially sequence the blank array below the original image matrix Appending to the top of the original image matrix, and after moving a column of blank arrays each time, step (2) is repeatedly performed to obtain (m+1)th to (mxn)th expansion matrices. The image pre-processing method according to claim 1, wherein the step (G) is a point multiplication of the temporary storage matrix and the second mask matrix by using a parallel computing technique. An image preprocessing method includes: (A) obtaining an original image matrix having a size of [MN]; (B) using the original image matrix as an initial matrix, and further obtaining a plurality of to-be-observed image matrices of the original image matrix as a function of time. (C) select a feature area in the original image matrix and Converting the feature region to a first mask matrix having a size of [mn]; (D) selecting a matrix of the region to be tested of [mn] size with the position of any element in the original image matrix as an origin, and then A blank array is added around the original image matrix to form mxn expansion matrices; (E) a temporary storage matrix having a size of [MxN mxn] is formed according to the size of the first mask matrix and the size of the original image matrix, wherein the element MxN For storing the position of the origin selected by the original image matrix, the element mxn is used to memorize the elements of the matrix of the area to be tested; (F) repeating step (D) and step (E) until the original image All elements in the matrix are calculated to fill the temporary storage matrix; (G) the first mask matrix corresponding to the temporary storage matrix is expanded into a second mask matrix having a size of [MxN mxn]; The temporary storage matrix is dot-multiplied with the second mask matrix to obtain a result matrix of [MxN mxn] size, and then the result matrix is gradually converted according to a predetermined rule to obtain a position of the feature region in the original image; And (G) taking each of the image frames to be observed Substituting the original image matrix, performing steps (D) to (H) to obtain the position of the feature region in each of the image matrix to be observed, and establishing a trajectory of the feature region moving with time. An image processing system includes: an original image module for acquiring an original image matrix having a size of [M N]; and an image processing algorithm module for using a preset image The calculation formula is converted into a first mask matrix having a size of [mn]; the expansion matrix module selects a matrix of the area to be tested of [mn] size by taking the position of any element in the original image matrix as an origin. Adding a blank array around the original image matrix to form mxn expansion matrices; the temporary storage matrix module is formed according to the size of the first mask matrix and the size of the original image matrix to form a size of [MxN mxn] a memory matrix, wherein the element MxN is used to store the position of the origin selected by the original image matrix, the element mxn is used to memorize the elements of the matrix of the area to be tested; and the loop processing module is used for repeated execution. The function of the expansion matrix module and the temporary storage matrix module until all elements in the original image matrix are calculated to fill the temporary storage matrix; the mask matrix processing module is configured to use the first mask matrix The temporary storage matrix is expanded into a second mask matrix having a size of [MxN mxn]; and the image processing module is configured to multiply the temporary storage matrix and the second mask matrix to obtain [MxN mxn The knot of size The matrix is then scaled according to a predetermined rule to obtain a new image matrix.