TW201025186A - Image processing method for providing depth information - Google Patents

Abstract

An image processing method is provided. The image processing method processes a number of capturing images, which are captured at a number of capturing angles, and thus produces a depth map of a viewing angle. By using a multi-resolution mesh established according to this invention, and calculating the depth value of the vertex of the multi-resolution mesh to obtain the depth value of other pixels, the depth map can be obtained. The multi-resolution mesh relates to a virtual viewing plane corresponding to the viewing angle. The multi-resolution mesh has a number of meshes having different resolutions. Each mesh has a number of vertexes. Each depth value of the vertex is generated according to at least a portion of the capturing images. During the establishment of the multi-resolution mesh, if the relationship of the depth values in the mesh having a first resolution conforms to a first predetermined condition, then the mesh, which has the first resolution and conforms to the first predetermined condition, is divided so as to generate a number of meshes having a second resolution.

Description

201025186 I WHO/^Γ/Λ. IX. Description of the Invention: [Technical Field] The present invention relates to an image processing method, and more particularly to an image processing method capable of providing a depth map. [Prior Art] In recent years, the stereoscopic image processing technology has been receiving increasing attention, and it can provide a stereoscopic image having a stereoscopic effect for the user to view. • Three-dimensional content to autostereoscopic display is also available to provide stereoscopic images. The above three-dimensional content includes a two-dimensional image and a three-information. This depth information corresponds, for example, to a depth map of one of the two-dimensional images. That is, the depth map contains depth values having respective elements corresponding to the two-dimensional image. According to the 2D image and the corresponding depth map, a multi-view stereo image can be displayed on the stereo display, so that the user can obtain the 3D viewing effect without the special lens ® . In the patent disclosed in U.S. Patent No. 5,361,127, the inventor uses a multi-faceted lens to capture images on different refractive faces. In this way, the depth value corresponding to the depth map of the photographed object can be obtained based on the images. Further, in the patent disclosed in U.S. Patent No. 6,487,304, a method and system are proposed which use multiple sets of images to obtain the depth of the depth map or to obtain the motion of the scene object 201025186 , ,

1 W48 UY£K style. In the patent disclosed in U.S. Patent No. 7,359,547, the inventor proposes to obtain the depth value of the depth map based on a plurality of images taken under various lighting conditions. Although a variety of depth map practices have been proposed, how to efficiently generate a depth map corresponding to a depth map of a two-dimensional image to reduce the amount of system computation is still one of the topics of the industry. SUMMARY OF THE INVENTION The present invention provides an image processing method for generating a depth map corresponding to a viewing angle based on a plurality of captured images captured by a plurality of image pickers located at different shooting positions. This method includes the following steps. First, create a set of initial meshes. The initial mesh is related to the virtual image plane corresponding to the viewing angle. The initial set of meshes has a plurality of meshes of a first resolution, and each of the nets of the first resolution has a plurality of tops 2 of the first resolution. Next, a depth value of the apex of the first resolution is generated based on at least a portion of the captured image. Determining whether a relationship of depth values of vertices in each of the first resolution grids meets a first predetermined condition, and if so, dividing a grid of first resolutions that meet the first predetermined condition to generate a plurality of second A grid of resolutions, which produces a grid of first resolutions and a grid of second resolutions (multi_res〇luti〇n^^). The grid of the second resolution each has a plurality of vertices of the second resolution. Thereafter, a depth value of the apex of the second resolution is generated based on at least a portion of the captured image. Then, the depth value of the apex of the first resolution and the depth value of the apex of the second resolution are used to generate a depth 6 201025186 i who value that does not correspond to the other pixels of the vertices to generate a depth map. In order to make the above description of the present invention more comprehensible, a preferred embodiment will be described below in detail with reference to the accompanying drawings. [Embodiment] The present invention provides an image processing method for generating a depth corresponding to a viewing angle according to a plurality of captured images captured by a plurality of image capturing devices located at different shooting positions. map. Referring to Figure 1, there is shown a flow chart of an image processing method in accordance with an embodiment of the present invention. This method includes the following steps. First, as shown in step S102, a set of initial meshes is created. This group of initial grids is related to one of the virtual image planes corresponding to the viewing angle. The initial grid of the set has a plurality of grids of a first resolution, and each grid of the first resolution has a plurality of vertex of the first resolution. Then, as shown in step S104, depth values of the vertices of the first resolutions are generated according to at least some of the captured images. Then, as shown in step ® S106, it is judged whether the relationship of the depth values of the vertices in the respective first resolution grids conforms to a first predetermined condition. If yes, proceeding to step S108, dividing a grid of the first resolution that meets the first predetermined condition to generate a plurality of grids of the second resolution, and generating a grid including the first resolutions and the second parsing A multi-resolution grid of one of the grids of degrees. The grids of the second resolution each have a plurality of vertices of the second resolution. Then, as shown in step S110, the depth values of the vertices of the second resolution are generated according to at least some of the captured images. Then, as step 7 201025186 '

Iw^e/zrA S112, according to the depth values of the vertices of the first resolution and the depth values of the vertices of the second resolutions, to generate depth values of other pixels that do not correspond to the vertices, To generate the depth map described above. The detailed description of the image processing method of Fig. 1 is explained below. Please refer to Figure 1, Figure 2A and Figure 3 at the same time. Figure 2A is a schematic diagram showing a portion of a set of initial meshes MRM0. Fig. 3 is a diagram showing an example of the relationship between the observation position Vp and a plurality of test depth planes TP1 to TPM. In step S102, a set of initial meshes MRM0 is created, as shown in Fig. 2A. The set of initial mesh MRM0 is associated with a virtual image plane 404 corresponding to the viewing angle of view 402. The virtual image plane 4〇4 is a plane substantially perpendicular to the direction in which the viewing angle 402 is viewed, as shown in Fig. 3. This set of initial mesh MRM0 has a plurality of meshes of a first resolution, such as mesh M1. Each first resolution grid has a plurality of vertices of a first resolution. For example, the grid M1 is a triangular grid, and the grid mi has vertices mla, mlb, and mlc. Then, as shown in step S104, the depth values of the vertices of the first resolution are generated according to at least some of the captured images. Please refer to FIG. 4, which is a flowchart of sub-steps sl4a, si04b and S104c included in step S104. As shown in Fig. 3, in step S104a, a plurality of predetermined depth values 沁 to < corresponding to the plurality of test depth planes TP1 to TPM are set. The scene in the space can be divided into a plurality of test depth planes TP1 TPPM, and the test depth planes TP1 TPPM are respectively a distance of 201025186 X ** TVi / Λ 1 from the observation position vp, the plurality of the above The distance system is related to a predetermined depth value. Since the human eye is more sensitive to the object at a close distance, the embodiment adopts a method in which the difference between the predetermined depth values of the two test depth planes which are closer to the observation position Vp is smaller. For example, the difference between the predetermined depth values 沁 and A of the test depth planes TP1 and TP2 in space may be less than the difference between the predetermined depth values A and A of the test depth planes T P 2 and T P 3 in the space. The predetermined depth value 沁~4 is set, for example, in the following manner: "m= 1 ^ m (1 (Expression Φ ^max ^min ^max where M is the number of predetermined depth values for this purpose. For the mth predetermined depth value (that is, the predetermined depth value of the test depth plane TPm), w is an integer between 1 and ;; i/max is the maximum value of the predetermined depth values. The minimum value of the predetermined depth values is expressed. According to Equation 1, the predetermined depth values are distributed in the disparity space, and the difference between the two predetermined depth values closer to the observation position Vp is smaller. According to the viewing angle of view 402, a part of the captured images are selected according to the viewing angle of view 402. The selected plurality of captured images are defined as a plurality of adjacent captured images. The method for selecting adjacent captured images is, for example, based on each image. For the distance between the shooting position of the picker and the direction of the viewing angle of view 402, the captured images corresponding to the shooting positions of the portion closer to the viewing angle of view 402 are selected as the adjacent captured images. ,to In the figure 3, when calculating the distance between each shooting position relative to the direction of the viewing angle of view 402, for example, the direction vector corresponding to the viewing angle of view 402 9 201025186 ·. I W48 先 is first found, and the shooting positions C1 to C5 are calculated and viewed. The vertical distances of the direction vectors corresponding to the viewing angle 402 are, for example, the vertical distances pdl to pd5 of the third figure. It is assumed that four shooting images corresponding to the four shooting positions having a close distance to the viewing angle 402 are selected. For the adjacent captured image, in this example, the selected adjacent captured image will be the captured image taken from the shooting positions C1 to C4. Then, as shown in step S104c, the vertex of each of the first resolutions is respectively Color Consistency Verification (CCV) is performed based on the image information of the adjacent captured images, and one of the predetermined depth values is selected as the depth value corresponding to the apex of the first resolution. For example, the apex m 1 a of Figure 2A is taken as an example. Please refer to Figure 3 and Figure 5 at the same time. Figure 5 shows the selected proximity photography. The schematic diagram of one case. Suppose the selected images of adjacent photographing

Iml to Im4 are captured images obtained at the shooting positions C1 to C4. This embodiment uses plane sweeping meth〇d to select one of these predetermined depth values as the depth ❹ value of the corresponding first resolution vertices. The so-called plane scanning method refers to first selecting one of a plurality of predetermined depth values for color-induced verification. If the selected predetermined depth value is correct or close, when the image block corresponding to the test depth plane is back-projected to a plurality of shooting positions, the color response of the generated image block will be the same. If they are inconsistent, then choose another predetermined depth value, and then do - color-caused verification. By performing color consistency verification by selecting test depth planes having different predetermined 201025186 a. ¥>«♦〇/zm depth values one by one, it is possible to find a predetermined depth value in which the color reaction is most consistent, as The depth value corresponding to the vertex of the first resolution. The practice of color consistency verification is further explained below. As shown in Fig. 5, first, one image block Pal to Pa4 is selected for each of the adjacent captured images Iml to Im4. The image blocks Pal to pa4 are the shadow image blocks corresponding to the respective adjacent captured images Iml to Im4 when the vertex mia is backprojected to the shooting positions C1 to C4 via the test depth plane TPi having the predetermined depth value d1. Then, the image blocks Pal~Pa4 of the adjacent captured images Iml~im4 are subjected to color consistency verification. In the implementation, the color consistency verification is performed on the image blocks Pal~Pa4 of the adjacent captured images Iml~Im4, for example, first, the image blocks Pal~Pa4 of any two adjacent captured images Iml~Im4 are obtained. The correlation coefficient of the color intensity (intensity). Each correlation coefficient is obtained according to the following formula: ν ν[Σ*(νί)2][Σ*σ;^ϊ (Formula 2) where 'π is the image block of the fth and _/th neighboring captured images The correlation coefficient of the intensity of the color, / is recognized as the color intensity of the A: the corresponding pixel in the / and y adjacent shots; 7; and is the color of the z and y adjacent shots respectively The average of the intensities. For example, it can be seen from the second formula that 〇2 is the correlation coefficient of the color intensities of the image blocks Pal and Pa2 adjacent to the captured images Iml and Im2. " 201025186 ,

1 W4872FA and heart are respectively the color intensity of the corresponding pixel in the image blocks Pal and Pa2 of the adjacent image Im 1 and Im2; f and 7; respectively, the image block Pal of the adjacent image Iml and Im2 The average of the color intensity of Pa2. At the same time, the similarity of the color intensity is determined by the similarity of the color images, such as Iml and Im3, imi and Im4, Im2 and Im3, Im2 and Im4, lm3 and Im4, respectively. And define the ccV value equal to 1 minus, ~, ~, ~, and the average value.如此 So repeatedly calculate a CCV value for each test depth plane with different predetermined depth values, and select the predetermined depth value corresponding to the smallest CCV value as the depth value corresponding to the apex of the first resolution. . After step S104, step S106 is performed to determine whether the relationship of the depth values of the vertices in the grids of the respective first resolutions meets a first predetermined condition. If yes, go to step S108 to divide the grid of the first resolution that meets the first predetermined condition to generate a plurality of grids of the second resolution. Preferably, in step S106, if the maximum value of the difference between the vertices of any two first resolutions of the grid of the first resolution is greater than a first predetermined value, the network of the first resolution The grid meets the first predetermined condition described above. Taking the grid M1 as an example, the first predetermined condition described above is explained as follows. Please refer to FIG. 2B, which is a schematic diagram showing an example of a grid M1 of the first resolution and a grid of the second resolution. In this example, whether the maximum value of the difference between the depth values of the two vertices of the judgment mesh M1 is greater than a first predetermined value is determined by 12 201025186 to determine whether the first predetermined condition is satisfied in the mesh M1. Any of the above two vertices refers to any two of the vertices mla to mlc, and is judged according to the following formula: max mD -ma p,qe{\X3}, p^g μ q >τ (Equation 3) %~% is represented as the second vertex of the mesh M1, that is, two of the vertices mla~mlc. Γ is the first predetermined value. In practice, the first predetermined value can be set depending on the situation. If the maximum value of the difference between the depth values of the two vertices is larger than the first predetermined value in Equation 3, that is, in the mesh M1, it means that the depth values corresponding to the three vertices of the mesh M1 vary greatly. At this time, the mesh M1 will be divided into three meshes of the second resolution, for example, the mesh M1'. The second resolution grid has three vertices mla', mlb', and mlc' of the second resolution. Then, as shown in step S110, the depth values of the vertices of the second resolution are generated according to at least some of the captured images. The method of finding the depth values of the vertices of the second resolutions can be obtained by using the depth values of the vertices of the first resolutions described above. Thus, after step S110, the initial mesh MRM0 will become the multi-resolution mesh MRM1 as shown in Fig. 2B. Please refer to FIG. 6 , which illustrates an example of a set of multi-resolution grids and corresponding three-dimensional stereo images of the embodiment. It is assumed that the three-dimensional image Im has a scene object Sp, and the scene object Sp is, for example, a sphere. This set of multiresolution meshes has multiple meshes of different resolutions, such as mesh M. The distribution of the grids with different resolutions will be compared with the three-dimensional image I m 13 201025186 seen from this viewing angle.

The scene object Sp of the TW4872FA is related. In this embodiment, the distribution of the meshes of different resolutions is determined according to the degree of change of the depth value of the scene object Sp of the three-dimensional image Im. For example, the scene object of the three-dimensional image Im includes, for example, the area A1 and the area A2, and the area A1 of the scene object near the boundary is another area. In the area A1, the variation of the depth value of the scene object is small, so the resolution of the grid is low; and in the area A2, the variation of the ice value of the % scene object Sp is large, so the grid The resolution is higher. That is, the area A1 has a smaller number of lower resolution grids and the area A2 has a larger number of grids with higher resolution. Therefore, it is known that the present embodiment can provide a mesh with a relatively high resolution (that is, a mesh with a higher resolution) on the boundary of the scene object, which will improve the accuracy of the depth map, and Can reduce the amount of calculations. Then, as shown in step S112, according to the depth values of the vertices of the first resolution and the depth values of the vertices of the second resolutions, to generate depth values of other pixels that do not correspond to the vertices, Generate the depth map described above. 4 The step S112 is described in detail below. In practice, step jg 112 may include the following steps. First, using the depth values of the vertices of the first resolution grid and the depth values of the vertices of the second resolutions, the set of multi-resolution grids MRM1 is vertex miing. Please refer to FIG. 7A and FIG. 7B. FIG. 7A shows an example of the group of multi-resolution grids MRM1 before the vertex filling process, and the 201025186 !· ▼» ΙΟ /▲!· 7Β One example of this group of multi-resolution grids MRM2 after performing vertex filling processing. In the seventh diagram, the mesh of the first resolution that does not meet the first predetermined condition (such as the mesh M2) is further divided to generate another mesh of the second resolution (such as the mesh M2, ). Taking the mesh M2 as an example, when the vertex filling process is performed, the vertices m2a to m2c' of the mesh M2 are used to obtain the midnight values of the other vertices mx and my of the second resolution. That is, the set of multi-resolution grids MRM1 is filled with the highest resolution vertices, and the filled grids are shown, for example, by the hollow dots of Figure 7B. Then, the depth of the vertex corresponding to the hollow dot can be obtained by linear interpolation using the vertex of the depth value that has been obtained (as indicated by the solid dot). For example, the vertex m2a and m2b are used to obtain the depth value of the filled apex mx. Thus, after the vertex filling process, a first-order multi-resolution mesh MRM2 is generated, and all the meshes of the set of multi-resolution mesh MRM2 are the mesh with the highest resolution. Then, the pixei interpolation process is performed on the set of multi-resolution mesh MRM2 after the vertex filling process to generate a depth map in which all the pixels have depth values. For example, in the embodiment, when the pixel interpolation processing is performed, the bilinear (bi_linear) interpolation method, that is, the horizontal interpolation and the vertical interpolation are simultaneously performed, to obtain corresponding to each The depth value of the pixels is such that a depth map corresponding to the depth values of each pixel can be generated. The following is an example of obtaining a depth value corresponding to a predetermined pixel. This predetermined pixel is a pixel that does not correspond to the vertices of the multi-resolution grid MRM2. Please refer to FIG. 7C, which is a schematic diagram of performing a pixel interpolation process on a predetermined pixel by using a bilinear interpolation 15 201025186 L W46 /Ζ1Ά complement method to obtain a deep production value thereof. The depth value of the predetermined element is obtained by the following formula: X f(u0, v0)= f(u% νΉΐ β )+f(u5+l, vs) α (1 - ^ } +f(u',ν' +1)(1—a)々+f(u,+1,ν,+1)α stone: (u,,ν′), (u,+ l,ν,), (u',ν, +1), and (u, +1, ν, +ι denote the coordinates of the adjacent four vertices; (uo, vo) is the coordinate of the predetermined gold; α denotes the coordinates of the predetermined pixel (u0, v0) The lateral distance of the vertex (u, v,); recorded as the longitudinal distance between the coordinates of the predetermined pixel (UG, VG) and the top 1 v value coordinate; and f ( ·) is expressed as the depth corresponding to the coordinate, please refer to Figure K5. Figure i shows a two-dimensional original image - an example. Figure 2 (d) is a depth map corresponding to the two-dimensional original image 1 degree grid MRM1. Figure 3 is continued to Figure 2 = The depth map of the generated divergence grid MRM2 after the vertex filling process is performed by the resolution grid MRM1. FIG. 4 shows the pixel interpolation process after the _^ re-resolution grid MRM2 is performed. Generated ^ Map DM. 'The degree of clothing and the calculation of depth values for all the elements to generate depth maps In the image processing method of the embodiment, all the pixel depth values are obtained by generating the depth values of the vertices of the multi-resolution grid. Thus, the data processing method of the image processing method of the embodiment is consistently It is lower, so the processing speed can be improved. As can be seen from Fig. 4, the depth map generated in this example is indeed very close to each of the books shown in Fig. 5 201025186 Λ. « * I \/ / ΛΛΜ Λ Λ is the unit' to find the depth map DM2 generated by the depth value of each pixel separately. Further, the image processing method proposed by the embodiment of the present invention can be a multi-core processor (muiicore pr〇cess) 〇r) is performed by means of parallel processing. For example, the multi-core processor can perform the above steps by using parallel processing, for example, grouping the grids in parallel to process 'by storing the related data. The multi-core processor can also perform parallel processing on the plurality of memory units. The multi-core processor can also perform the vertex filling processing, or the pixel interpolation processing, and the like described above in a parallel processing manner to generate the depth map DM. This will further improve the processing speed of the image processing method of the embodiment. In another example, the step of generating the depth map dm may further include the following steps: providing depth values of the vertices of the grids To one

A Graphic Processing Unit (GPU), and thus the GPU uses its internal rendering function to generate this depth map. The processing speed of the image processing method of the present embodiment can be further improved by using a GPU having a function of converting the depth value of the vertices of the multi-resolution grid MRM1 described above into the depth map DM. In addition, the depth map generated by the image processing method of the present embodiment, after combining the two-dimensional composite scene corresponding to the viewing angle 402, can be used to generate stereoscopic images of other viewing angles. The above embodiment is described by taking a grid having a first resolution and a second resolution as a set of multiple resolution grids, but is not limited thereto. This set of multi-resolution grids can also be a set of multiples with N resolutions. 17 201025186

1 W4»/^A resolution grid. The value of N here can be set according to the situation. For example, the method of generating a grid of third resolution is as follows. First, the grid of the corresponding second resolution is selectively segmented according to the relationship of the depth values of the vertices in the grid of the second resolution to generate a plurality of grids of the third resolution. The third resolution grids each have a plurality of vertices of a third resolution. Then, the depth values of the vertices of the third resolution are generated based on at least some of the captured images. Therefore, it can be inferred that according to the relationship between the depth values of the vertices in the third resolution, the corresponding third resolution mesh can be further divided to generate a plurality of meshes having higher resolution. Thus, after calculating the depth values of the vertices of the respective meshes, a set of multiple resolutions having N kinds of resolutions can be generated. Moreover, the image processing method disclosed in the above embodiments of the present invention is configured to process a plurality of captured images in a plurality of shooting angles to generate a depth map in a viewing angle. By establishing a set of multi-resolution grids and calculating the depth values of the vertices of the set of multi-resolution grids, you can quickly find the depth values of other pixels to quickly obtain this depth map. The invention has the advantages of high efficiency, fast processing speed, and good accuracy of the depth map. In view of the above, although the present invention has been disclosed above in a preferred embodiment, it is not intended to be used in comparison with the present invention. This (4) is a person of ordinary skill in the technical field, and can make various changes and refinements without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention is defined by the scope of the appended patent. . Month 18 201025186 X TT -T«J / ΓΛ. [Simple Description of the Drawings] FIG. 1 is a flow chart showing an image processing method according to an embodiment of the present invention. Figure 2A is a schematic representation of a portion of a set of initial meshes. FIG. 2B is a schematic diagram showing an example of a grid of the first resolution and a grid of the second resolution. Figure 3 shows an example of the relationship between the observation position and a plurality of test depth planes. ❹ FIG. 4 is a flow chart showing sub-steps S104a, S104b and S104c included in step S104. Fig. 5 is a schematic view showing an example of the selected adjacent captured image. Figure 6 is a diagram showing an example of a set of multi-resolution grids and corresponding three-dimensional images of the present embodiment. Fig. 7A is a diagram showing an example of the set of multiple resolution grids MRM1 before the vertex filling process. ® 7B shows an example of this set of multi-resolution grid MRM2 after vertex filling processing. Fig. 7C is a schematic diagram showing the pixel value interpolation process of a predetermined pixel by using a bilinear interpolation method to obtain the depth value thereof. [Main component symbol description] C1~C5: Shooting position DM: Depth map 201025186

TW4872PA

Im : 3D image

Iml~Im4: adjacent shooting images

Imo: two-dimensional original image mla~mlc, mla'~mlc', m2a~m2c: vertex

Ml, ΜΓ, M2: Grid MRM0: Initial Grid MRM1: Multiple Resolution Grid

Pal~Pa4: image block S102~S112, S104a~S104c: process steps ❿

Sp : Scene object TP1 ~ TPM: Test depth plane

Vp: observation position 402: viewing angle 404: virtual image plane

20 201025186 [Description of the Drawings] Figure 1 shows an example of a two-dimensional original image. 2 is a depth map of a multi-resolution grid MRM1 corresponding to a two-dimensional original image. 3 is a depth map of the multi-resolution grid MRM2 generated after the vertex filling process of the multi-resolution grid MRM1 of FIG. 2 is performed. FIG. 4 is a diagram showing the depth map generated after the multi-resolution grid MRM2 of FIG. 3 is subjected to the pixel interpolation processing. Figure 5 is a diagram showing the depth map generated by finding the depth value of each element in each pixel.

A

Claims (1)

201025186 i YY-tot^r^k X. Patent application scope: 1. An image processing method, ::r=r, two:== 1 depth view of one of the viewing angles, the method includes · (4) Constructing a set of initial meshes 'The initial mesh system of the set is related to the pseudo-image plane of the viewing angle'. The initial mesh system has a plurality of "grids". Each of the first-resolution meshes has a plurality of a vertex of the first resolution; generating a depth value of the vertices of the first resolution according to at least some of the captured images; determining a relationship of depth values of the vertices in each of the first-resolution grids疋No--the predetermined condition, if yes, dividing the grid of the first-resolution corresponding to the first predetermined condition to generate a plurality of grids of the second resolution and generating the first-resolution a plurality of resolution grids of the grid and the second resolution grids, the second resolution grids each having a plurality of second resolution vertices; according to at least some of the shots The image produces the depth of the vertices of the second resolution a depth value; and a depth value of the vertices of the second resolutions, and depth values of the vertices of the second resolutions to generate depth values of other pixels that do not correspond to the vertices to generate the depth 2. The method of claim 1, further comprising: selectively segmenting the correspondence according to a 21 201025186 1 W4tt/^A relationship of depth values of vertices in a second resolution grid a grid of the second resolution to generate a plurality of third resolution grids, each of the third resolution grids having a plurality of third resolution vertices; and according to at least some of the The method of shooting the image to generate the depth value of the vertices of the third resolution. 3. The method of claim 1, wherein the step of generating the depth values of the vertices of the first resolution is used. 4. The method according to claim 1, wherein in the step of dividing the grid corresponding to the first resolution, if a grid of the first resolution is Any two of these The maximum value of the difference between the vertices of the resolution is greater than a first predetermined value, and the grid of the first resolution meets the first predetermined condition. 5. The method of claim 1, wherein the method is generated The step of the depth map includes: performing vertex filling on the set of multi-resolution grids by using the depth values of the vertices of the first resolution and the depth values of the vertices of the second resolutions. Processing; and performing pixel interpolation processing on the set of multi-resolution grids after vertex filling processing to generate the depth map in which all pixels have depth values. 6. The method of claim 5, wherein the step of performing a vertex fill process on the set of multi-resolution grids comprises: dividing the web of the first resolution that does not meet the first predetermined condition 22 201025186 1 » ** TU / ^ 1 / - L · grid to generate another plurality of second resolution grid resolution grid filled with the apex of the second resolution 7 · as claimed in the sixth item The method described in which the vertex fill processing of the set of re-resolution grids is performed by a linear lion insertion method, and the vertices of the first and other grids that have obtained the depth values are used to obtain other The depth value of the other vertices of the first mesh is obtained. Φ [As in the method of claim 5, in which the pixel interpolation method of the vertice St group multi-resolution grid is performed by the interpolation method of the gold and the bilinear (bMlnear) to obtain the corresponding The method of claim 2, wherein the step of using the double= interpolation method to obtain the depth value corresponding to each element further comprises: obtaining the corresponding to-predetermined The depth value of the pixel, the predetermined pixel is a pair of pixels of the set of multi-resolution grids after the vertex fill processing, and the depth value of the predetermined pixel is obtained by the following formula: f(uO, V〇)= f(U',V')(11 )(1—none)+f(u,+l,V,)α (1 - /5 ) +%',v'+l ) (1— a)々+f(u,+1,ν,+ 1)α stone: (u,v ), (u'+1,v'), (u',v'+1), And (u, +i, v, +l) are expressed as coordinates of the adjacent four vertices; 23 201025186 1 w*to/^r/v (uo, vo) is the coordinate of the predetermined pixel; α is expressed as The coordinates of the predetermined element (uo, vo) and the vertices (u,, the lateral distance of the coordinates; ' > 泠 denotes the coordinate of the predetermined pixel (uo, vo) and the vertex (u, the longitudinal distance of the coordinates; and ' v) f ( ·) is expressed as the depth value corresponding to the coordinate. The method of claim 1, wherein the step of generating depth values for vertices of the first resolution includes: Setting a plurality of predetermined strict values corresponding to the plurality of test depth planes; 'clothing degree according to the viewing angle, selecting a part of the captured images, and the captured images are defined as a plurality of adjacent captured images; For each of the vertices of the first resolutions, performing color-based verification based on the image information of the adjacent photos, and selecting one of the predetermined depth values as the corresponding vertex of the first resolution Second degree. The method of claim 10, wherein the step of capturing the image adjacent to the image comprises: selecting, according to a distance between a shooting position of each image picker and a direction of the viewing angle; The photographed image corresponding to the photographing position of the portion of the viewing angle that is closer to the viewing angle, as the method of the first aspect of the patent application, wherein the method is The multi-core processor is implemented using parallel processing 24 201025186. The method of claim 1, wherein the step of generating the depth map comprises: providing the depth values of the vertices of the grids to a graphics processing unit (GPU); And the depth map is generated by the GPU using its internal rendering function. 25