TW201910584A - Patch embroidering method based on image recognition which can automatically embroider an applique of any outer frame pattern without being limited to the outer frame pattern - Google Patents

Abstract

The invention provides a patch embroidering method based on image recognition. First, an image of a fabric to be embroidered is captured by a photographing box, wherein the fabric to be embroidered is attached to a base fabric fixed to a bottom fabric of an embroidery frame. A computing device is used to correct the image and calculate a plurality of smoothed edge points of the fabric to be embroidered in the image, then the invention obtains an universal coordinates corresponding to each edge point based on the image coordination and the thickness correction. Finally, the computerized sewing machine automatically embroider the fabric according to an embroidery path formed by the universal coordinates corresponding to the edge points. The invention can make the computerized sewing machine automatically embroider the applique of any outer frame pattern, which is not limited to the outer frame pattern pre-stored in the computerized sewing machine.

Description

Appliqué method based on image recognition

The invention relates to a method for appliqué embroidering, in particular to a method for appliqué embroidering based on image recognition.

In the conventional method of automatically embroidering the appliqué by using a computer-type sewing machine, the computer-type sewing machine needs to pre-store the outer frame pattern of the appliqué, for example, a circular shape or an animal shape; the user needs to store in advance Make a selection in the frame pattern, and then let the computerized sewing machine automatically embroider the appliqué of the selected frame pattern.

For example, as shown in FIG. 1(a), the user selects a circular pattern from the computer-type sewing machine, and causes the computer-type sewing machine to automatically embroider the circular pattern 1 on the cloth member 21 fixed to the embroidery frame 23; Next, as shown in FIG. 1(b), the user cuts the circular pattern 1 from the cloth member 21 and sticks it to the base fabric 22 fixed to the embroidery frame 23, and then causes the computer-type sewing machine to automatically apply the circle. The outer frame of the pattern 1 is embroidered.

However, it is far from sufficient for the user to allow the computerized sewing machine to embroider the patch of the pre-stored frame pattern.

SUMMARY OF THE INVENTION Accordingly, it is an object of the present invention to provide an image recognition-based appliqué method which allows a computerized sewing machine to automatically embroider an applique of any frame pattern.

Therefore, the patch embedding method based on image recognition of the present invention is implemented by an embroidery system and is used for embroidering a fabric to be embroidered attached to a base fabric. The embroidery system corresponds to a world coordinate system, and the image recognition based patch embedding method comprises a step (a), a step (b), a step (c), and a step (d).

The step (a) is to capture a first image of the fabric to be embroidered, wherein the first image corresponds to an image coordinate system.

The step (b) is to perform edge detection on the first image to calculate a plurality of edge points of the fabric to be embroidered in the first image.

The step (c) is to calculate a world coordinate corresponding to each edge point in the world coordinate system according to the image coordinate system and the world coordinate system.

The step (d) is to embroider the member to be embroidered according to an embroidery path formed by the world coordinates corresponding to the edge points.

The effect of the invention is that the computerized sewing machine can automatically embroider the appliqué of any outer frame pattern.

The patch embedding method based on image recognition of the present invention is implemented by an embroidery system comprising a conventional photographing box, a computing device, and a conventional computerized sewing machine. The computing device is, for example, a desktop computer or a tablet computer. Preferably, the camera lens of the camera box is a wide-angle lens or a fisheye lens.

Referring to Fig. 2, an embodiment of the patch embedding method based on image recognition of the present invention will be described in detail below.

First, in step 31, as shown in FIG. 3 and FIG. 4, the user attaches a fabric to be embroidered 41 to a base fabric 42 fixed to an embroidery frame 43, wherein the outer frame pattern of the fabric to be embroidered 41 can be It is an arbitrary shape, and is described here in a love shape. Then, the cassette 431 of the embroidery frame 43 is pivotally connected to the cassette 53 of the photographing box 5 to fix the embroidery frame 43 in the photograph box 5, and the photographing box 5 is used to capture the to-embroidered member 41. a first image.

Next, in step 32, in order to correct the distortion of the first image, the first image is image-corrected by the computing device to obtain a first corrected image. The present invention also provides an innovative image correction method, the details of which will be explained at the end of this section.

Then, in step 33, referring to FIG. 5, the edge correction algorithm, such as a Canny edge detector, is used by the computing device to obtain the image to be embroidered in the first corrected image 64. A plurality of edge points E _{1 of the} cloth member 41 are further subjected to a smoothing operation on the edge points to obtain a plurality of other edge points. Preferably, the smoothing operation is as follows: Referring to FIG. 5 and FIG. 6, the edge points E ₁ obtained by the edge detection algorithm are first used to estimate a boundary point E ₁ . Parametric non-uniform rational B-spline curve (NURBS curve), and then sample a plurality of edge points E ₂ from the parametric non-uniform rational B-spline curve Cr, the edges The point E ₂ is the edge point after the smoothing operation. However, it should be specially noted that if the edge detection points E ₁ detected by the edge detection algorithm have no obvious burrs/saw edges, the above smoothing operation is unnecessary. Further, the manner of the smoothing operation is not limited to the above, and for example, a smoothing operation may be performed by a general smoothing filter.

Then, in step 34, using the computing device, each edge point E ₂ is calculated in the world coordinate system according to an image coordinate system corresponding to the first corrected image and a world coordinate system corresponding to the computerized sewing machine. The corresponding world coordinates, that is, the so-called image alignment.

In detail, referring to FIG. 7 and FIG. 8, the computerized sewing machine 7 includes a working platform 71 and a moving module 72. The mobile module 72 has a body 721 and a connecting unit 722 connected to the body 721 and movable along the body 721. After the embroidery frame 43 is connected to the connecting unit 722 through the cassette 431, the body 721 is The connecting unit 722 can move the embroidery frame 43 in the x-axis direction and the y-axis direction of the world coordinate system WCS corresponding to the computer-type sewing machine 7 to perform embroidery. The image is aligned in the following manner. First, as shown in Fig. 8, a cloth member 8 is fixed to the embroidery frame 43, and the computerized sewing machine 7 is operated to embroider a plane coordinate system PCS on the cloth member 8. Next, the embroidery frame 43 is placed in the photographing box 5 and a second image of the plane coordinate system PCS is captured by the photographing box 5. Then, the second image is corrected to generate a second corrected image. Next, referring to FIG. 9, by superimposing the first corrected image 64 and the second corrected image (not shown), the plane coordinate system PCS in the second corrected image is calculated relative to the first A rotation matrix and a displacement vector of the image coordinate system ICS corresponding to the corrected image 64. Then, according to the rotation matrix, the displacement vector, and a pair of image ratios of the first corrected image 64 and the second corrected image, each edge point E _{2 of} the member to be embroidered 41 is The coordinate on the image coordinate system ICS is converted into a world coordinate in the world coordinate system WCS, wherein the unit of the image ratio is, for example, "micrometer/pixel", "mm/pixel", "cm/pixel", etc. That is, the length in the real space corresponding to each pixel of the first corrected image 64 or the second corrected image. Further, let c _i denote a coordinate of an edge point E ₂ on the image coordinate system ICS, c _w denotes a world coordinate corresponding to c _i on the world coordinate system WCS, and A and b respectively represent the rotation matrix and The displacement vector, and α is the image scale, then c _w = α( Ac _i +b ).

Next, in step 35, as shown in FIG. 10, the embroidery frame 43 holding the fabric to be embroidered 41 is connected to the moving module 72 of the computerized sewing machine 7, so that the computerized sewing machine 7 can be based on the edge points E. ₂ corresponding to a path of the embroidery coordinate world those formed, the pieces of cloth to be embroidered embroidery 41; thus, the embroidery can be obtained the results shown in FIG. 11.

In particular, the above-described image recognition-based appliqué method allows the computer-type sewing machine 7 to automatically embroider the appliqué of any frame pattern prepared by the user, and is no longer limited to being pre-stored in the computer-type sewing machine 7. The frame of the frame.

In an embodiment, if the thickness of the fabric to be embroidered 41 or the base fabric 42 is large, the world coordinates corresponding to the edge points E ₂ obtained in the step 34 may be before the step 35. Perform "cloth thickness correction" as detailed below. Referring to FIG. 12, since the lens 51 of the photographing box 5 captures an image, not only the incident light H ₁ in the vertical direction but also the incident light H ₂ in the non-vertical direction, each of the first corrected images 64 The linear distance between the world coordinates corresponding to the edge point E ₂ and the origin O of the world coordinate system is too large, so the distance needs to be reduced, and if the thickness of the base fabric 42 and the thickness of the fabric to be embroidered 41 are greater, Large, the larger the reduction will be.

For example, as shown in FIG. 12, it is assumed that the thickness of the base fabric 42 is T ₀ , the thickness of the fabric to be embroidered 41 is T ₁ , the distance from the lens 51 of the photographing box 5 to the bottom 52 thereof is D, and after the first correction image to be embroidered cloth member 64 the edge of a _two pairs of points E 41 should be a member of embroidered cloth actual edge point P _t 41, then at step 34 that the edge point calculated P _t The distance between the world coordinates and the origin O of the world coordinate system is L ₁ , and the distance L ₁ needs to be reduced to the actual distance L _{11 of} the edge point P _t to the origin O of the world coordinate system. The similar triangle theorem can calculate the magnitude of the reduction as L _{1 -} L ₁₁ = L ₁ ÷ D × (T ₀ + T ₁ ). Similarly, the distance between the world coordinate of the actual edge point Q _t of the fabric to be embroidered 41 calculated by the step 34 and the origin O of the world coordinate system is L ₂ , and the distance L ₂ needs to be reduced to The edge point Q _{t is} to the actual distance L ₂₁ of the origin O of the world coordinate system, and the magnitude of the reduction can be calculated by the similar triangle theorem as L _{2 -} L ₂₁ = L ₂ ÷ D × T ₀ .

In an embodiment, the thickness of the world coordinates corresponding to each edge point E ₂ in the first corrected image 64 may not be corrected. For example, if the computing device determines that the distance between the world coordinate corresponding to an edge point E ₂ and the origin O of the world coordinate system is less than a threshold value, the world coordinates corresponding to the edge point E ₂ are not thickened. Correction; and when the distance is greater than the threshold value, the thickness of the world coordinates corresponding to the edge point E _{2 is} corrected.

The present invention provides an innovative image correction method, which is described in more detail below. Referring to FIGS. 13-15, first, in step 321, a calibration plate is placed in the photo box 5 and an image of the calibration plate is captured by the photo box 5. Preferably, the correction plate is a checkerboard correction plate 61 as shown in FIG.

Next, in step 322, the plurality of image feature points of the image of the calibration plate are captured by the computing device. Preferably, as shown in FIG. 15, a plurality of corner points 621 of the image 62 of the checkerboard correction plate are captured by Harris corner detection as the image feature points, wherein the images are The feature point is a floating point.

Next, in step 323, the computing device calculates a geometric surface having a plurality of control parameters and fitting the image feature points according to the image feature points. Preferably, as shown in FIG. 16, the geometric curved surface is a parametric non-uniform rational B-spline surface (NURBS surface), wherein the conventional parametric NURBS surface is utilized. Parametric NURBS surface interpolation to estimate the parametric non-uniform rational B-spline surface 63 fitting the image feature points by using the image feature points as interpolation points, that is, Where {W _i,j } is the weight value, and {P _i,j } is a plurality of control points 631 (control parameters) that are also floating point numbers calculated using the image feature points, u and v are The two axial directions of the domain of the parametric non-uniform rational B-spline surface 63 [0,1], v [0,1], {N _i,p (u)} and {N _j,q (v)} are both B-spline basis functions, and p and q are the u-axis directions, respectively. The degree with the direction of the v-axis.

Next, in step 324, the parametric non-uniform rational B-spline curved surface 63 is used to correct a to-be-corrected image captured by the photographing box 5 to generate a corrected image. For convenience of explanation, the following is an image 62 of the checkerboard correction plate shown in FIG. 15 as the image to be corrected.

In detail, first, as shown in FIG. 17, a first pixel number of a first image axis (x-axis) of the corrected image 68 and a second image axis of the corrected image 68 are defined ( A second number of pixels of the y-axis); here, the number of the first pixels is k and the number of the second pixels is t.

Next, referring to FIG. 17 to FIG. 19, the first pixel number of sampling points {u _i |i=1, 2 is defined on the u-axis of the parametric non-uniform rational B-spline surface. ..., k}, and defining the second pixel number of sampling points {v _j |j=1, 2,..., t} on the v-axis of the parametric non-uniform rational B-spline surface And the plurality of curved surface points on the parametric non-uniform rational B-spline curved surface 63 calculated according to the sampling points are in one-to-one correspondence with the pixels of the corrected image 68; wherein, preferably, the The distance between any two adjacent sampling points on the u-axis is the same, that is, 1/k, and the distance between any two adjacent sampling points on the v-axis is the same, that is, 1/t, and (u _i , v _j The surface point corresponding to the parametric non-uniform rational B-spline surface 63 is S((i-0.5)/k, (j-0.5)/t). That is, assuming f(i,j) represents the (i,j)th pixel of the corrected image 68, then f(i,j) corresponds to (u _i ,v _j ) and the surface point S((i -0.5) / k, (j - 0.5) / t), where i is a positive integer from 1 to k and j is a positive integer from 1 to t. That is, as shown in FIG. 18, the definition field 65 is divided into a plurality of equal squares 651 having the same number of pixels as the corrected image 68, and the squares 651 and the corrected squares 651 The pixels of the image 68 correspond one-to-one, and the surface points corresponding to each pixel are the surface points corresponding to the center point of the square 651 corresponding to the pixel. In addition, referring to FIG. 19, each square 651 of the definition field 65 corresponds to a polygonal area 652 on the parametric non-uniform rational B-spline surface 63, and each polygon area 652 includes a pair of images 68 that should be corrected. A pixel's surface point 653.

Then, for each pixel f(i,j) of the corrected image 68, the surface point 653 corresponding to the pixel f(i,j) is used to interpolate at least one adjacent pixel in the image to be corrected 69. And calculating a pixel value of the pixel f(i, j); wherein the interpolation manner may be, for example, a conventional nearest neighbor interpolation, a bilinear interpolation, or a higher Sub-interpolation and other methods. For example, for the pixel f(5, 6) of the corrected image 68, the pixel value is S(4.5/k, 5.5/t) corresponding to (u ₅ , v ₆ ) in the to be corrected. The at least one neighboring pixel in the image 69 is interpolated and calculated.

Referring to FIG. 20, it is worth mentioning that, since each surface point is a floating point number, if the image to be corrected 69 has M×N pixels, the image coordinate corresponding to the image to be corrected 69 needs to cover C1 (- Four endpoints: 0.5, -0.5), C2 (M-1+0.5, -0.5), C3 (M-1+0.5, N-1+0.5), C4 (-0.5, N-1+0.5) Defining a target plane 9 to cover a surface point located at a boundary of the parametric non-uniform rational B-spline surface; and the center of the (i, j)th pixel of the image to be corrected 69 corresponds to the image coordinate system The coordinates are (i-1, j-1), where i is a positive integer from 1 to M, and j is a positive integer from 1 to N.

In addition, referring to FIG. 21, in another embodiment, a weighted average of pixel values of all pixels covered by the polygon region 652 including S (4.5/k, 5.5/t) in the image to be corrected 69 may also be calculated. (weighted mean), the pixel value of the pixel f(5, 6) of the corrected image 68; wherein the weight of each pixel is the area ratio of the pixel in the polygonal area 652. For example, as shown in FIG. 21, the polygon area 652 covers the pixel portion P ₁ of the area of A _1, the pixel portion P ₂ of the area of A _2, an area of the pixel portion P _3, A _3, P ₄ of the pixel section area of a _4, part of the pixel region P of the area a _{5 5;} order , the weighted average is ,among them Is the weight of the pixel P _i . In another embodiment, the weights may be defined in the pixel P _i weight, wherein the shorter the distance the greater the weights according to the distance from the center S (4.5 / k, 5.5 / t) of the pixel P _i.

In particular, by using the curved surface points 653, any image captured by the photographing box 5 can be corrected, including the first image and the second image. Referring to FIG. 22, FIG. 22 illustrates the corrected image 68 obtained by correcting the image 69 to be corrected, that is, the image 62 of the checkerboard.

In particular, the image correction method described above can correct image distortion caused by the geometric design of the lens of the camera lens, and can correct the distortion of the lens manufacturing, the lens assembly position is not accurate, and the image sensor assembly of the camera is assembled. Image distortion caused by factors such as inaccurate position. In addition, the deformation of the subject itself can be corrected and flattened in the image.

In addition, in addition to correcting distortion distortion of the image to be corrected, the image correction method may further set a number of pixels of the first image axis (x-axis) of the corrected image and the second image axis (y The number of pixels of the axis) sets the image size of the corrected image.

In addition, the calibration plate can also be in other implementations. For example, the calibration plate may be a dot correction plate 67 as shown in FIG. 23, and the image of the dot correction plate 67 is image-recognized by the calculation device, and the center of each dot is extracted and utilized. The point center is used as the image feature points; and other implementation steps are as described above, and are not described here.

In addition, although the image correction method of the present invention is preferably used to perform image correction on the first image and the second image, the image is not limited thereto. For example, a pinhole camera model can be used. And performing image correction on the first image to generate the first corrected image, and obtaining a world coordinate corresponding to each edge point of the fabric to be embroidered in the first corrected image.

In addition, it is worth mentioning that if the distortion of the image captured by the camera box 5 is relatively slight, the image correction performed on the first image and the second image is not necessary.

In summary, the present invention is based on an image-applied patch embedding method, by capturing an image of the fabric to be embroidered, and performing image correction on the image and calculating a plurality of smoothed portions of the fabric to be embroidered. Processing the edge points, and obtaining the world coordinates corresponding to each edge point through image alignment and cloth thickness correction, and then causing the computerized sewing machine to automatically follow the embroidery path formed by the world coordinates corresponding to the edge points. The embroidery to be embroidered is embroidered, so that the object of the present invention can be achieved.

However, the above is only the embodiment of the present invention, and the scope of the invention is not limited thereto, and all the equivalent equivalent changes and modifications according to the scope of the patent application and the patent specification of the present invention are still The scope of the invention is covered.

1‧‧‧Circular pattern

21‧‧‧ cloth

22‧‧‧Backcloth

23‧‧‧Embroidery frame

31~35‧‧‧Steps

321~324‧‧‧Steps

41‧‧‧Embroidery pieces to be embroidered

42‧‧‧Backcloth

43‧‧‧Embroidery frame

431‧‧‧Carmen

5‧‧‧Photo Box

51‧‧‧ lens

52‧‧‧ bottom

53‧‧‧Carmen

61‧‧‧checkerboard correction board

62‧‧‧Image of checkerboard correction board

621‧‧‧ corner

63‧‧‧Parametric non-uniform rational B-spline surface

631‧‧‧Control points

64‧‧‧First corrected image

65‧‧‧Parameter-type non-uniform rational B-spline surface

651‧‧‧ square

652‧‧‧Polygonal area

653‧‧‧Surface points

67‧‧‧Mark Calibration Board

68‧‧‧Fixed image

69‧‧‧Image to be corrected

7‧‧‧Computerized sewing machine

71‧‧‧Working platform

72‧‧‧Mobile Module

721‧‧‧Ontology

722‧‧‧ Connection unit

8‧‧‧ cloth

9‧‧‧ coordinate plane

f(i,j)‧‧‧ the (i,j)th pixel of the corrected image

C ₁ ~ C ₄ ‧ ‧ four endpoints of the image coordinate plane

P ₁ ~P ₅ ‧‧ ‧ pixels

Part area of A ₁ ~A ₅ ‧‧ ‧ pixels

Cr‧‧‧Parametric non-uniform rational B-spline curve

E ₁ ‧‧‧Image edge points before smoothing

E ₂ ‧‧‧Image edge points after smoothing

WCS‧‧‧World coordinate system

ICS‧‧‧image coordinate system

PCS‧‧‧planar coordinate system

H ₁ ‧‧‧ incident light in the vertical direction

H ₂ ‧‧‧non-vertical incident light

P _t ‧‧‧The actual edge point of the fabric to be embroidered

Q _t ‧‧‧Another actual edge point of the fabric to be embroidered

O‧‧‧ origin of the world coordinate system

Other features and advantages of the present invention will be apparent from the embodiments of the drawings, wherein: Figure 1 illustrates a conventional manner of making a patchwork using a computerized sewing machine; Figure 2 illustrates the image recognition based on the present invention. Fig. 3 shows a method of attaching a fabric to be embroidered to a base fabric fixed to an embroidery frame; Fig. 4 is a schematic view of a first image of the fabric to be embroidered by using a photographing box; a plurality of edge points of the fabric to be embroidered before the smoothing operation in a first corrected image; FIG. 6 is a schematic diagram of the smoothing operation of the fabric to be embroidered in the first corrected image after FIG. Figure 7 illustrates a computerized sewing machine; Figure 8 illustrates the computerized sewing machine embroidering a plane coordinate system on a cloth member; Figure 9 illustrates the relative coordinate of the plane coordinate system with an image coordinate system in the image. Figure 10 illustrates the embroidering of the fabric to be embroidered according to an embroidery path formed by the plurality of world coordinates corresponding to the edge points of the computer-type sewing machine; Figure 11 is a schematic illustration of the appliqué after the embroidery is completed; Figure 12 illustrates the use of this FIG. 13 illustrates an image correction method of the present invention; FIG. 14 illustrates a checkerboard correction plate; FIG. 15 illustrates an image of the checkerboard correction plate and a plurality of corner points thereof. Figure 16 illustrates a parametric non-uniform rational B-spline surface estimated from the isocenter and its plurality of control points; Figure 17 illustrates the number of pixels of a corrected image corresponding to a to-be corrected image; Figure 18 illustrates The parametric non-uniform rational B-spline surface defines the domain; FIG. 19 illustrates the pixel value of each pixel of the corrected image with reference to FIG. 17 and FIG. 18; FIG. 20 illustrates an image coordinate system required to cover Figure 21 illustrates a calculation of a pixel value of the corrected image; Figure 22 illustrates the corrected image obtained by correcting the image of the checkerboard; and Figure 23 illustrates a dot Calibration board.

Claims (10)

An appliqué method based on image recognition, implemented by an embroidery system, and used for embroidering a fabric to be embroidered attached to a base fabric, the embroidery system corresponding to a world coordinate system, the image recognition based patch The embedding method comprises the following steps: (a) capturing a first image of the fabric to be embroidered, wherein the first image corresponds to an image coordinate system; (b) performing edge detection on the first image to calculate a plurality of edge points of the fabric to be embroidered in the first image; (c) calculating, according to the image coordinate system and the world coordinate system, a world coordinate corresponding to each edge point in the world coordinate system; and (d) And embroidering the article to be embroidered according to an embroidery path formed by the world coordinates corresponding to the edge points. The image recognition-based appliqué method according to claim 1, wherein the step (b) comprises the following sub-steps: (b1) obtaining an image to be embroidered in the first image by using an edge detection algorithm a plurality of edge points; and (b2) performing a smoothing operation on the edge points to obtain a plurality of edge points. The image recognition-based appliqué method according to claim 2, wherein in the step (b2), the edge points obtained in the step (b1) are first used to obtain a fitting edge point. The parametric non-uniform rational B-spline curve is sampled from the parametric non-uniform rational B-spline curve. The image recognition-based appliqué method according to claim 1, wherein the step (c) comprises the following sub-steps: (c1) embroidering a pair of world coordinate systems on an object based on the world coordinate system a plane coordinate system; (c2) capturing a second image of the plane coordinate system; (c3) calculating the plane coordinate system in the second image by superimposing the first image and the second image a rotation matrix and a displacement vector of the image coordinate system; and (c4) according to the rotation matrix, the displacement vector, and a pair of image ratios of the first image and the second image, each edge point is at the image coordinate The coordinates on the system are converted into a world coordinate in the world coordinate system. The image recognition-based applique embroidery method according to claim 1, further comprising a step (e) performed after the step (c) and performed before the step (d): corresponding to at least one edge point a world coordinate, wherein at least one of a thickness of the base fabric and a thickness of the fabric to be embroidered reduces a distance of a world coordinate corresponding to each edge point to an origin of the world coordinate system, wherein the reduced distance is positively correlated At least one of a thickness of the base fabric and a thickness of the fabric to be embroidered. The image recognition-based appliqué method according to claim 1, further comprising a step (f) performed after the step (a) and before the step (b): performing image correction on the first image . The image recognition-based appliqué method according to claim 6, wherein the step (f) comprises the following sub-steps: (f1) capturing an image of a calibration plate; (f2) capturing a plurality of the image Image feature points; (f3) calculating a geometric surface having a plurality of control parameters and fitting the image feature points according to the image feature points; and (f4) generating a pair according to the first image and the geometric surface a first corrected image of the first image, wherein each pixel of the first corrected image corresponds to a curved point on the geometric surface, and the pixel value of each pixel is based on the curved surface corresponding to the pixel at the first At least one neighboring pixel in the image is calculated. The image recognition-based appliqué method according to claim 7, wherein in the step (f3), the geometric surface is a parametric non-uniform rational B-spline surface, and each control parameter is non-uniform A control point for a rational B-spline surface. The image recognition-based appliqué method according to claim 8, wherein the step (f4) comprises the following sub-steps: (f41) defining a first pixel number of a first image axis of the first corrected image And defining a second number of pixels of a second image axis of the first corrected image; and (f42) defining the first number of pixels on a first domain axis of the domain of the geometric surface Sampling a point, and defining a second number of sampling points on a second domain axis of the domain of the geometric surface, wherein the plurality of surface points on the geometric surface are calculated according to the sampling points The pixels of the first corrected image are in one-to-one correspondence. The image recognition-based applique embroidery method according to claim 9, wherein in the step (f42), the distance between any two adjacent sampling points on the first domain axis of the domain of the geometric surface is the same, The distance between any two adjacent sampling points on the second domain of the domain of the geometric surface is the same, and the pixel value of each pixel of the first corrected image is in the first image by using the corresponding surface point At least one neighboring pixel is calculated by one of interpolation and weighted averaging.