WO2021000471A1 - 一种高分辨率图像匹配方法及系统 - Google Patents
一种高分辨率图像匹配方法及系统 Download PDFInfo
- Publication number
- WO2021000471A1 WO2021000471A1 PCT/CN2019/114525 CN2019114525W WO2021000471A1 WO 2021000471 A1 WO2021000471 A1 WO 2021000471A1 CN 2019114525 W CN2019114525 W CN 2019114525W WO 2021000471 A1 WO2021000471 A1 WO 2021000471A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- matching
- resolution image
- pixel
- image
- resolution
- Prior art date
Links
- 238000000034 method Methods 0.000 title claims abstract description 70
- 239000000523 sample Substances 0.000 claims abstract description 59
- 238000004364 calculation method Methods 0.000 claims abstract description 39
- 230000008569 process Effects 0.000 claims abstract description 29
- 238000005070 sampling Methods 0.000 claims description 43
- 238000012545 processing Methods 0.000 claims description 9
- 238000012790 confirmation Methods 0.000 claims description 6
- 125000004122 cyclic group Chemical group 0.000 claims description 3
- 238000010586 diagram Methods 0.000 description 8
- 230000008447 perception Effects 0.000 description 7
- 238000005457 optimization Methods 0.000 description 5
- 230000000007 visual effect Effects 0.000 description 4
- 238000013461 design Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000003190 augmentative effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/75—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
- G06V10/751—Comparing pixel values or logical combinations thereof, or feature values having positional relevance, e.g. template matching
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/761—Proximity, similarity or dissimilarity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/75—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
- G06V10/759—Region-based matching
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
Definitions
- the invention belongs to the technical field of image processing, and particularly relates to a high-resolution image matching method and system.
- Vision-based depth perception is a general-purpose underlying support technology that is widely used in many fields such as autonomous driving, industrial inspection, robotics, augmented reality, and drones.
- visual depth perception has the advantages of obtaining dense data, high accuracy, low cost, and good applicability.
- due to the problem of matching between images visual depth perception also has the problem of large amount of calculation, and the application in many fields that generally require real-time response faces big problems.
- it in order to ensure real-time performance, it can generally only support the matching of low-resolution images. As a result, the performance of current high-resolution camera equipment cannot be fully utilized, and only depth data with lower accuracy and smaller distance can be obtained .
- image matching for visual depth perception mainly includes local methods and global methods.
- the local method uses pixel-by-pixel comparison to find pixels corresponding to the same target in two images to form a match between two image pixels. Therefore, the corresponding disparity and depth data are calculated.
- This method faces a large number of pixel comparison calculations and requires a large amount of calculation. In order to ensure real-time performance, it can only support low-resolution image matching.
- the global method transforms the matching of images into an optimization problem, and realizes the matching between images by finding the optimal solution in the global scope. Since global optimization involves the design of a large number of parameters, and the solution of the optimal solution is very complicated and unstable, although this method is better than the local method, it also faces more complicated calculation processes.
- This method uses pixel-by-pixel comparison to achieve image matching calculations like the local method, but optimizes the parallax results in the corresponding directions through optimization methods in several directions. Improved the accuracy of the parallax value.
- the semi-global optimization method improves to a certain extent the problems of low accuracy of local methods and complex calculations of global methods, but it does not completely solve the problems of high-resolution image matching and long-distance depth perception.
- the present invention provides a method that can not only reduce the calculation amount of the image matching process, but also improve the accuracy of the matching result. Based on the overall consistency, the matching of high-resolution images is obtained through reverse refinement. The resulting high-resolution image matching method.
- the technical solution provided by the present invention is: a high-resolution image matching method, which includes the following steps:
- the obtained matching result of the low-resolution image is reverse refined to obtain the matching results of the high-resolution images at various levels until the matching result of the initial resolution image is obtained.
- the step of performing regional fidelity down-sampling on the initial high-resolution image to obtain a multi-level low-resolution image specifically includes the following steps:
- the above-mentioned down-sampling process is performed in sequence until the resolution of the obtained image meets the sampling stop condition, the sampling process is stopped, and the finally obtained image is the image with the lowest resolution.
- the information of four adjacent pixels up, down, left, and right is integrated to obtain a value, and each obtained value contains the information of four adjacent images.
- the step of performing local matching on the obtained multi-level low-resolution image with a global probe, and obtaining the matching result of the low-resolution image specifically includes the following steps:
- the determined number, direction, and length of the probes calculate the size relationship between the pixel value and the center pixel value of the lowest resolution image at each position on the probe, which is expressed as a binary value;
- each pixel to be matched For each pixel to be matched, perform a weighted summation according to the matching cost corresponding to each matching result obtained by calculation and the distance between the corresponding probes obtained, and the weighted sum result obtained is used as the final matching cost of each pixel to be matched;
- the candidate pixel with the smallest matching cost is selected as the final matching result of the low-resolution image.
- the overall consistency of image matching is used to reverse refine the matching results of the obtained low-resolution images to obtain the matching results of high-resolution images at various levels until the initial resolution is obtained.
- the steps of matching results of rate images specifically include the following steps:
- the matching calculation of the optimal pixel matching relationship in the pixel area and the determining operation steps of the matching relationship between the current resolution image are cyclically executed until the matching result of the highest resolution image is obtained as the final initial resolution image matching result.
- Another object of the present invention is to provide a high-resolution image matching system, which includes:
- the down-sampling module is used to down-sample the initial high-resolution image with regional fidelity to obtain a multi-level low-resolution image
- the local matching module is used to perform local matching on the obtained multi-level low-resolution image with a global probe to obtain a matching result of the low-resolution image;
- the reverse refinement module is used to use the overall consistency of image matching to perform reverse refinement on the matching results of the obtained low-resolution images to obtain the matching results of high-resolution images at all levels until the initial resolution is obtained The matching result of the image.
- the down-sampling module specifically includes:
- the value initialization module is used to initialize the value of the number of pixels R, and the number of pixels R is used as a sampling stop condition;
- the pixel value confirmation module is used to use the value of the obtained pixel number R as the pixel value of the corresponding position of the next-level resolution image;
- the lowest-resolution image acquisition module is used to sequentially execute the above-mentioned down-sampling process until the obtained image resolution meets the sampling stop condition, stop the sampling process, and the finally obtained image is the image with the lowest resolution.
- the information of four adjacent pixels up, down, left, and right is integrated to obtain a value, and each obtained value contains the information of four adjacent images.
- the local matching module specifically includes:
- Probe initialization module used to initialize the number, direction and length of global probes
- the pixel matching module is configured to perform pixel-by-pixel matching on the obtained lowest resolution image, and calculate the matching cost corresponding to each matching result;
- the size relationship calculation module is used to calculate the size relationship between the pixel value and the center pixel value of the lowest resolution image at each position on the probe according to the determined number, direction and length of the probe, expressed in binary value;
- Probe distance calculation module used to calculate the distance between corresponding probes for all pixels to be matched
- the final matching cost calculation module is used to perform weighted summation for each pixel to be matched according to the matching cost corresponding to each matching result calculated and the distance between the corresponding probes obtained, and the weighted sum result obtained is used as each pixel to be matched.
- the final matching result confirmation module is used to select the candidate pixel with the smallest matching cost as the final matching result of the low-resolution image.
- the reverse refinement module specifically includes:
- the pairing processing module is used to perform pairing processing on each matching pixel in the obtained matching result of the low-resolution image, and find its respective pixel area in the upper-level resolution image;
- the optimal pixel matching relationship calculation module is used to perform matching calculation of pixels in the corresponding pixel area in the obtained upper-level resolution image to obtain the optimal pixel matching relationship in the pixel area;
- a matching relationship determination module configured to determine the obtained optimal pixel matching relationship in the pixel area as the matching relationship between the current resolution images
- the cyclic execution module is used to cyclically execute the matching calculation of the optimal pixel matching relationship in the pixel area and the determination of the matching relationship between the current resolution image until the matching result of the highest resolution image is obtained as the final initial resolution image Match results.
- the initial high-resolution image is down-sampled with regional fidelity to obtain a multi-level low-resolution image; the obtained multi-level low-resolution image is locally performed with a global probe.
- Figure 1 is a flowchart of the implementation of the high-resolution image matching method provided by the present invention
- Fig. 2 is a flow chart of implementing regional fidelity down-sampling of an initial high-resolution image provided by the present invention to obtain a multi-level low-resolution image;
- FIG. 3 is a flow chart of the implementation provided by the present invention for performing local matching on the obtained multi-level low-resolution image with a global probe to obtain the matching result of the low-resolution image;
- Figure 4 is the use of the overall consistency of image matching provided by the present invention to reverse refine the matching results of the obtained low-resolution images to obtain the matching results of high-resolution images at various levels until the initial resolution image is obtained Flow chart of the realization of matching results;
- Figure 5 is a schematic diagram of the process of regional fidelity downsampling provided by the present invention.
- Fig. 6 is a schematic diagram of a region matching process with a global probe provided by the present invention.
- FIG. 7 is a schematic diagram of a matching result in a low-resolution image provided by the present invention.
- Figure 8 is a structural block diagram of the high-resolution image matching system provided by the present invention.
- Figure 9 is a structural block diagram of the down-sampling module provided by the present invention.
- Figure 10 is a structural block diagram of a local matching module provided by the present invention.
- Fig. 11 is a structural block diagram of a reverse refinement module provided by the present invention.
- Fig. 1 shows the implementation flow chart of the high-resolution image matching method provided by the present invention, which specifically includes the following steps:
- step S101 area fidelity down-sampling is performed on the initial high-resolution image to obtain a multi-level low-resolution image.
- step S102 the obtained multi-level low-resolution image is locally matched with a global probe to obtain a matching result of the low-resolution image.
- step S103 using the overall consistency of the image matching, the obtained matching result of the low-resolution image is reverse refined to obtain the matching results of the high-resolution images at various levels until the matching of the initial resolution image is obtained. Knot.
- a regional fidelity down-sampling strategy is designed to convert high-resolution images into low-resolution images while maintaining more image information.
- An image matching method with global probes is proposed, which uses global probes to optimize the matching accuracy and reduce mismatches caused by local similarity.
- a reverse refinement method based on overall consistency is designed to optimize the matching process of low-resolution image matching results to high-resolution image matching results, and further improve the matching accuracy of high-resolution images.
- Fig. 2 shows the implementation flow chart of performing regional fidelity down-sampling on the initial high-resolution image provided by the present invention to obtain a multi-level low-resolution image, which specifically includes the following steps:
- step S201 the value of the number of pixels R is initialized, and the number of pixels R is used as a sampling stop condition.
- step S202 the value of the number of pixels R obtained is used as the pixel value of the corresponding position of the next-level resolution image.
- step S203 the above-mentioned down-sampling process is sequentially performed until the obtained image resolution meets the sampling stop condition, the sampling process is stopped, and the finally obtained image is the image with the lowest resolution.
- the information of the four adjacent pixels up, down, left, and right is synthesized to obtain a value, and each obtained value contains the information of four adjacent images to ensure the local area
- the information can be held in the next level of pixels.
- the regional fidelity down-sampling strategy converts high-resolution images into low-resolution images while maintaining more image information.
- Fig. 3 shows a flow chart of implementing the local matching of the obtained multi-level low-resolution image by means of a global probe provided by the present invention to obtain the matching result of the low-resolution image.
- the steps specifically include The following steps:
- step S301 initialize the number, direction and length of the global probes
- the global probes can be four-way, eight-way, etc., and the number is optional.
- the direction can be aligned along the axis, diagonal, or any other direction.
- the length of the probe can extend to the edge of the image or to the middle of the image.
- step S302 perform pixel-by-pixel matching on the obtained lowest resolution image, and calculate the matching cost corresponding to each matching result;
- the selection of the matching window can be a regular shape, an adaptive irregular shape or any other shape.
- the matching cost can be directly calculated based on the pixel value, or any method such as non-parametric transformation based on the pixel value can be used.
- step S303 according to the determined number, direction and length of the probes, the size relationship between the pixel value and the center pixel value of the lowest resolution image at each position on the probe is calculated, expressed in binary values;
- the size relationship between all pixels on each probe and the center pixel forms a binary string according to the pixel position.
- step S304 for all pixels to be matched, the distance between the corresponding probes is calculated;
- the distance between the probes can be calculated using Hamming distance or any other distance calculation method.
- step S305 for each pixel to be matched, a weighted summation is performed according to the matching cost corresponding to each matching result calculated and the distance between the corresponding probes, and the weighted summation result is used as the result of each pixel to be matched.
- Final matching cost for each pixel to be matched, a weighted summation is performed according to the matching cost corresponding to each matching result calculated and the distance between the corresponding probes, and the weighted summation result is used as the result of each pixel to be matched.
- step S306 the candidate pixel with the smallest matching cost is selected as the final matching result of the low-resolution image.
- a high-precision local matching method with a global probe is designed.
- the optimal matching pair is jointly determined to avoid Fall into mismatches caused by local similarities.
- Fig. 4 shows the overall consistency of image matching provided by the present invention, the matching results of the low-resolution images obtained are reverse refined to obtain the matching results of high-resolution images at various levels until the initial resolution is obtained.
- the flow chart of the realization of the matching result of the rate image specifically includes the following steps:
- step S401 in the obtained matching result of the low-resolution image, a pairing process is performed on each matched pixel to find its respective pixel area in the upper-level resolution image;
- the correspondence between pixels of images with different levels of resolution can maintain corresponding data in the down-sampling process of step 1, and the correspondence data can also be obtained in other ways.
- step S402 for the corresponding pixel area in the obtained upper-level resolution image, the matching calculation of the pixels in the area is performed to obtain the optimal pixel matching relationship in the pixel area;
- the matching calculation can use the same method as the matching calculation in step 2, or any other method.
- the optimal pixel matching relationship in the pixel area is obtained.
- step S403 the obtained optimal pixel matching relationship in the pixel area is determined as the matching relationship between the current resolution images
- This process completes the transition from low-resolution image pixel matching relationship to high-resolution image pixel matching relationship, instead of directly using the pixel correspondence relationship between different resolution images as the matching result of the high-resolution image.
- step S404 the matching calculation of the optimal pixel matching relationship in the pixel area and the determining the matching relationship between the current resolution image are cyclically executed until the matching result of the highest resolution image is obtained as the final initial resolution image matching result.
- the present invention designs a reverse refinement method based on overall consistency, based on the matching result of the low-resolution image and the overall image matching. Consistency, through local refinement of the area where the pixel is located, the matching relationship corresponding to the high-resolution image is obtained.
- FIG. 8 shows a structural block diagram of the high-resolution image matching system provided by the present invention. For ease of description, only the parts related to the embodiment of the present invention are shown in the figure.
- the high-resolution image matching system includes:
- the down-sampling module 11 is used to perform regional fidelity down-sampling on the initial high-resolution image to obtain a multi-level low-resolution image;
- the local matching module 12 is configured to perform local matching on the obtained multi-level low-resolution image in a manner with a global probe to obtain a matching result of the low-resolution image;
- the reverse refinement module 13 is used to use the overall consistency of image matching to perform reverse refinement on the obtained matching results of the low-resolution images to obtain the matching results of high-resolution images at various levels until the initial resolution is obtained Rate the matching result of the image.
- the down-sampling module 11 specifically includes:
- the value initialization module 14 is used to initialize the value of the number of pixels R, and the number of pixels R is used as a sampling stop condition;
- the pixel value confirmation module 15 is configured to use the value of the number of pixels R obtained as the pixel value of the corresponding position of the next-level resolution image;
- the lowest-resolution image acquisition module 16 is configured to sequentially execute the above-mentioned down-sampling process until the obtained image resolution meets the sampling stop condition, stop the sampling process, and the finally obtained image is the image with the lowest resolution.
- the local matching module 12 specifically includes:
- Probe initialization module 17 used to initialize the number, direction and length of the global probe
- the pixel matching module 18 is configured to perform pixel-by-pixel matching on the obtained lowest resolution image, and calculate the matching cost corresponding to each matching result;
- the size relationship calculation module 19 is used to calculate the size relationship between the pixel value and the center pixel value of the lowest resolution image at each position on the probe according to the determined number, direction and length of the probe, expressed in binary value;
- the probe distance calculation module 20 is used to calculate the distance between the corresponding probes for all pixels to be matched;
- the final matching cost calculation module 21 is used to perform a weighted summation for each pixel to be matched based on the matching cost corresponding to each matching result calculated and the distance between the corresponding probes obtained, and the weighted sum result obtained is used as each The final matching cost of the pixels to be matched;
- the final matching result confirmation module 22 is configured to select the candidate pixel with the smallest matching cost as the final matching result of the low-resolution image.
- the reverse refinement module 13 specifically includes:
- the pairing processing module 23 is configured to perform pairing processing on each matched pixel in the obtained matching result of the low-resolution image, and find its respective pixel area in the upper-level resolution image;
- the optimal pixel matching relationship calculation module 24 is configured to perform matching calculation of pixels in the corresponding pixel region in the obtained upper-level resolution image to obtain the optimal pixel matching relationship in the pixel region;
- the matching relationship determination module 25 is configured to determine the obtained optimal pixel matching relationship in the pixel area as the matching relationship between the current resolution images;
- the cyclic execution module 26 is used to cyclically execute the matching calculation of the optimal pixel matching relationship in the pixel area and the determining operation steps of the matching relationship between the current resolution image, until the matching result of the highest resolution image is obtained as the final initial resolution Image matching result.
- the initial high-resolution image is down-sampled with regional fidelity to obtain a multi-level low-resolution image; the obtained multi-level low-resolution image is locally performed with a global probe.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Evolutionary Computation (AREA)
- Databases & Information Systems (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
- Microscoopes, Condenser (AREA)
Abstract
一种高分辨率图像匹配方法及系统，方法包括：对初始高分辨率图像进行区域保真下采样，得到多层次的低分辨率图像；采用带有全局探针的方式对得到的多层次的所述低分辨率图像进行局部匹配，得到低分辨率图像的匹配结果；利用图像匹配的整体一致性，对得到的所述低分辨率图像的匹配结果进行反向求精，得到各级高分辨率图像的匹配结果，直至得到初始分辨率图像的匹配结果，从而既能降低图像匹配过程的计算量，又能提高匹配结果的精度，再基于整体一致性，通过反向求精得到高分辨率图像的匹配结果。
Description
本申请要求于2019年6月29日提交中国专利局、申请号为201910580849.3、发明名称为“一种高分辨率图像匹配方法及系统”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
本发明属于图像处理技术领域，尤其涉及一种高分辨率图像匹配方法及系统。
基于视觉的深度感知是一种在自动驾驶、工业检测、机器人、增强现实、无人机等许多领域有着普遍应用的底层通用支撑技术。相较于其他深度感知手段，视觉深度感知具有可以得到稠密数据、精度高、成本低、适用性好等优势。但由于涉及到图像之间的匹配问题，视觉深度感知也存在计算量大的问题，在普遍要求实时响应的诸多领域的应用面临较大问题。当前视觉深度感知技术的应用中，为保证实时性普遍只能支持低分辨率图像的匹配，导致无法充分利用目前高分辨率摄像设备的性能，只能得到较低精度和较小距离的深度数据。
目前，用于视觉深度感知的图像匹配主要有局部方法和全局方法两种，局部方法通过逐像素比较的方式，在两幅图像中寻找对应同一目标的像素，形成两幅图像像素之间的匹配关系，从而计算对应的视差和深度数据；这种方法面临大量的像素比对计算，运算量大，为保证实时性只能支持低分辨率图像的匹配。全局方法把图像的匹配转化为一个优化问题，通过在全局范围内寻找最优解实现图像之间的匹配。由于全局优化涉及到大量参数的设计，且最优解的求解非常复杂和不稳定，这种方法的效果虽然比局部方法更好，但也面临计算过程更为复杂的问题。结合这两种方法各自的特点，出现了半全局优化方法，这种方法和局部方法一样采用逐像素比对实现图像的匹配计算，但是在若干方向上通过优化手段得到对应方向的优化视差结果，提高了视差值的准确性。半全局优化方法在一定程度上改善了 局部方法准确性不高和全局方法计算复杂的问题，但并未彻底解决高分辨率图像的匹配和远距离深度感知问题。
发明内容
针对现有技术中的缺陷，本发明提供了一种既能降低图像匹配过程的计算量，又能提高匹配结果的精度，再基于整体一致性，通过反向求精得到高分辨率图像的匹配结果的高分辨率图像匹配方法。
本发明所提供的技术方案是：一种高分辨率图像匹配方法，所述方法包括下述步骤：
对初始高分辨率图像进行区域保真下采样，得到多层次的低分辨率图像；
采用带有全局探针的方式对得到的多层次的所述低分辨率图像进行局部匹配，得到低分辨率图像的匹配结果；
利用图像匹配的整体一致性，对得到的所述低分辨率图像的匹配结果进行反向求精，得到各级高分辨率图像的匹配结果，直至得到初始分辨率图像的匹配结果。
作为一种改进的方案，所述对初始高分辨率图像进行区域保真下采样，得到多层次的低分辨率图像的步骤具体包括下述步骤：
初始化像素个数R的值，所述像素个数R作为停止采样条件；
将得到像素个数R的值作为下一级分辨率图像对应位置的像素值；
依次执行上述下采样过程，直到得到的图像分辨率满足采样停止条件，停止采样过程，最后得到的图像即为最低分辨率的图像。
作为一种改进的方案，在初始高分辨率图像过程中，综合上下左右相邻的四个像素信息，得到一个值，得到的每个值包含了相邻四个图像的信息。
作为一种改进的方案，所述采用带有全局探针的方式对得到的多层次的所述低分辨率图像进行局部匹配，得到低分辨率图像的匹配结果的步骤具体包括下述步骤：
初始化全局探针数量、方向和长度；
对得到的所述最低分辨率图像进行逐像素匹配，计算得到各个匹配结果对应的匹配代价；
根据确定的探针数量、方向和长度，计算所述最低分辨率图像在探针上各个位置处的像素值和中心像素值的大小关系，以二进制值表示；
对所有待匹配像素，计算其对应的探针之间的距离；
对每个待匹配像素，依据计算得到的各个匹配结果对应的匹配代价和得到的对应的探针之间距离进行加权求和，得到的加权求和结果作为各个待匹配像素的最终匹配代价；
选取匹配代价最小的候选像素作为所述低分辨率图像的最终匹配结果。
作为一种改进的方案，所述利用图像匹配的整体一致性，对得到的所述低分辨率图像的匹配结果进行反向求精，得到各级高分辨率图像的匹配结果，直至得到初始分辨率图像的匹配结果的步骤具体包括下述步骤：
在得到的低分辨率图像的匹配结果中，对每个匹配像素执行配对处理，找到其各自在上一级分辨率图像中对应的像素区域；
对得到的上一级分辨率图像中对应的像素区域，进行区域内像素的匹配计算，得到像素区域内最优像素匹配关系；
将得到的所述像素区域内最优像素匹配关系确定为当前分辨率图像之间的匹配关系；
循环执行像素区域内最优像素匹配关系的匹配计算和当前分辨率图像之间的匹配关系确定操作步骤，直到得到最高分辨率图像的匹配结果，作为最终的初始分辨率图像匹配结果。
本发明的另一目的在于提供一种高分辨率图像匹配系统，所述系统包括：
下采样模块，用于对初始高分辨率图像进行区域保真下采样，得到多层次的低分辨率图像；
局部匹配模块，用于采用带有全局探针的方式对得到的多层次的所述低分辨率图像进行局部匹配，得到低分辨率图像的匹配结果；
反向求精模块，用于利用图像匹配的整体一致性，对得到的所述低分 辨率图像的匹配结果进行反向求精，得到各级高分辨率图像的匹配结果，直至得到初始分辨率图像的匹配结果。
作为一种改进的方案，所述下采样模块具体包括：
值初始化模块，用于初始化像素个数R的值，所述像素个数R作为停止采样条件；
像素值确认模块，用于将得到像素个数R的值作为下一级分辨率图像对应位置的像素值；
最低分辨率图像获取模块，用于依次执行上述下采样过程，直到得到的图像分辨率满足采样停止条件，停止采样过程，最后得到的图像即为最低分辨率的图像。
作为一种改进的方案，在初始高分辨率图像过程中，综合上下左右相邻的四个像素信息，得到一个值，得到的每个值包含了相邻四个图像的信息。
作为一种改进的方案，所述局部匹配模块具体包括：
探针初始化模块，用于初始化全局探针数量、方向和长度；
像素匹配模块，用于对得到的所述最低分辨率图像进行逐像素匹配，计算得到各个匹配结果对应的匹配代价；
大小关系计算模块，用于根据确定的探针数量、方向和长度，计算所述最低分辨率图像在探针上各个位置处的像素值和中心像素值的大小关系，以二进制值表示；
探针距离计算模块，用于对所有待匹配像素，计算其对应的探针之间的距离；
最终匹配代价计算模块，用于对每个待匹配像素，依据计算得到的各个匹配结果对应的匹配代价和得到的对应的探针之间距离进行加权求和，得到的加权求和结果作为各个待匹配像素的最终匹配代价；
最终匹配结果确认模块，用于选取匹配代价最小的候选像素作为所述低分辨率图像的最终匹配结果。
作为一种改进的方案，所述反向求精模块具体包括：
配对处理模块，用于在得到的低分辨率图像的匹配结果中，对每个匹 配像素执行配对处理，找到其各自在上一级分辨率图像中对应的像素区域；
最优像素匹配关系计算模块，用于对得到的上一级分辨率图像中对应的像素区域，进行区域内像素的匹配计算，得到像素区域内最优像素匹配关系；
匹配关系确定模块，用于将得到的所述像素区域内最优像素匹配关系确定为当前分辨率图像之间的匹配关系；
循环执行模块，用于循环执行像素区域内最优像素匹配关系的匹配计算和当前分辨率图像之间的匹配关系确定操作步骤，直到得到最高分辨率图像的匹配结果，作为最终的初始分辨率图像匹配结果。
在本发明实施例中，对初始高分辨率图像进行区域保真下采样，得到多层次的低分辨率图像；采用带有全局探针的方式对得到的多层次的所述低分辨率图像进行局部匹配，得到低分辨率图像的匹配结果；利用图像匹配的整体一致性，对得到的所述低分辨率图像的匹配结果进行反向求精，得到各级高分辨率图像的匹配结果，直至得到初始分辨率图像的匹配结果，从而既能降低图像匹配过程的计算量，又能提高匹配结果的精度，再基于整体一致性，通过反向求精得到高分辨率图像的匹配结果。
为了更清楚地说明本发明具体实施方式或现有技术中的技术方案，下面将对具体实施方式或现有技术描述中所需要使用的附图作简单地介绍。在所有附图中，类似的元件或部分一般由类似的附图标记标识。附图中，各元件或部分并不一定按照实际的比例绘制。
图1是本发明提供的高分辨率图像匹配方法的实现流程图；
图2是本发明提供的对初始高分辨率图像进行区域保真下采样，得到多层次的低分辨率图像的实现流程图；
图3是本发明提供的采用带有全局探针的方式对得到的多层次的所述低分辨率图像进行局部匹配，得到低分辨率图像的匹配结果的实现流程图；
图4是本发明提供的利用图像匹配的整体一致性，对得到的所述低分辨率图像的匹配结果进行反向求精，得到各级高分辨率图像的匹配结果， 直至得到初始分辨率图像的匹配结果的实现流程图；
图5是本发明提供的区域保真下采样的过程示意图；
图6是本发明提供的带有全局探针的区域匹配过程示意图；
图7是本发明提供的低分辨率图像中匹配结果示意图；
图8是本发明提供的高分辨率图像匹配系统的结构框图；
图9是本发明提供的下采样模块的结构框图；
图10是本发明提供的局部匹配模块的结构框图；
图11是本发明提供的反向求精模块的结构框图。
下面将结合附图对本发明技术方案的实施例进行详细的描述。以下实施例仅用于更加清楚地说明本发明的、技术方案，因此只作为示例，而不能以此来限制本发明的保护范围。
图1示出了本发明提供的高分辨率图像匹配方法的实现流程图，其具体包括下述步骤：
在步骤S101中，对初始高分辨率图像进行区域保真下采样，得到多层次的低分辨率图像。
在步骤S102中，采用带有全局探针的方式对得到的多层次的所述低分辨率图像进行局部匹配，得到低分辨率图像的匹配结果。
在步骤S103中，利用图像匹配的整体一致性，对得到的所述低分辨率图像的匹配结果进行反向求精，得到各级高分辨率图像的匹配结果，直至得到初始分辨率图像的匹配结。
在该实施例中，设计了区域保真的下采样策略，将高分辨率图像转化为低分辨率图像，同时保持更多图像信息。提出了带有全局探针的图像匹配方法，利用全局探针优化匹配精度，减少由局部相似造成的误匹配。设计了基于整体一致性的反向求精方法，优化低分辨率图像匹配结果向高分辨率图像匹配结果对应过程，进一步提高高分辨率图像的匹配精度。
图2示出了本发明提供的对初始高分辨率图像进行区域保真下采样，得到多层次的低分辨率图像的实现流程图，其具体包括下述步骤：
在步骤S201中，初始化像素个数R的值，所述像素个数R作为停止采样条件。
在步骤S202中，将得到像素个数R的值作为下一级分辨率图像对应位置的像素值。
在步骤S203中，依次执行上述下采样过程，直到得到的图像分辨率满足采样停止条件，停止采样过程，最后得到的图像即为最低分辨率的图像。
其中，在该实施例中，在初始高分辨率图像过程中，综合上下左右相邻的四个像素信息，得到一个值，得到的每个值包含了相邻四个图像的信息，保证局部区域信息能够保持在下一级的像素中。
在该实施例中，如图5所示，区域保真的下采样策略，将高分辨率图像转化为低分辨率图像，同时保持更多图像信息。
图3示出了本发明提供的采用带有全局探针的方式对得到的多层次的所述低分辨率图像进行局部匹配，得到低分辨率图像的匹配结果的实现流程图，其步骤具体包括下述步骤：
在步骤S301中，初始化全局探针数量、方向和长度；
在该步骤中，全局探针可以是四向、八向等等，数量任选。方向可以是沿轴对齐方向，也可以是对角线方向，或者任意其它方向。探针长度可以延伸到图像边缘，也可以延伸到图像中间。
在步骤S302中，对得到的所述最低分辨率图像进行逐像素匹配，计算得到各个匹配结果对应的匹配代价；
在该步骤中，匹配窗口的选择可以是规则形状，也可以是自适应的不规则形状或其它任何形状。匹配代价可以基于像素值直接计算，也可以采用基于像素值的非参数变换等任何方法。
在步骤S303中，根据确定的探针数量、方向和长度，计算所述最低分辨率图像在探针上各个位置处的像素值和中心像素值的大小关系，以二进制值表示；
每个探针上所有像素和中心像素之间的大小关系值依像素位置形成一个二进制字符串。
在步骤S304中，对所有待匹配像素，计算其对应的探针之间的距离；
在该步骤中，探针之间距离的计算可以采用汉明距离，也可以采用任何其他距离计算方式。
在步骤S305中，对每个待匹配像素，依据计算得到的各个匹配结果对应的匹配代价和得到的对应的探针之间距离进行加权求和，得到的加权求和结果作为各个待匹配像素的最终匹配代价；
在步骤S306中，选取匹配代价最小的候选像素作为所述低分辨率图像的最终匹配结果。
在该实施例中，如图6所示，设计了一种带有全局探针的高精度局部匹配方法，通过计算当前匹配的局域匹配度和全局匹配度，共同决定最优匹配对，避免陷入因局部相似造成的误匹配。
图4示出了本发明提供的利用图像匹配的整体一致性，对得到的所述低分辨率图像的匹配结果进行反向求精，得到各级高分辨率图像的匹配结果，直至得到初始分辨率图像的匹配结果的实现流程图，其具体包括下述步骤：
在步骤S401中，在得到的低分辨率图像的匹配结果中，对每个匹配像素执行配对处理，找到其各自在上一级分辨率图像中对应的像素区域；
在该步骤中，不同级分辨率图像之间像素的对应关系可以在步骤一的下采样过程中维护相应数据，也可以采用其他方式获得该对应关系数据。
在步骤S402中，对得到的上一级分辨率图像中对应的像素区域，进行区域内像素的匹配计算，得到像素区域内最优像素匹配关系；
在该步骤中，该匹配计算可以采用和步骤二中匹配计算相同的方法，也可以采用其他任何方法。通过匹配计算，得到像素区域内最优的像素匹配关系。
在步骤S403中，将得到的所述像素区域内最优像素匹配关系确定为当前分辨率图像之间的匹配关系；
该过程完成了从低分辨率图像像素匹配关系向高分辨率图像像素匹配关系的过渡，而不是直接用不同分辨率图像间像素的对应关系作为高分辨率图像的匹配结果。
在步骤S404中，循环执行像素区域内最优像素匹配关系的匹配计算和 当前分辨率图像之间的匹配关系确定操作步骤，直到得到最高分辨率图像的匹配结果，作为最终的初始分辨率图像匹配结果。
在该实施例中，如图7所示，针对采样过程可能造成的引入误差，本发明设计了基于整体一致性的反向求精方法，基于低分辨率图像的匹配结果和图像匹配中的整体一致性，通过对像素所在区域进行局部求精，得到高分辨率图像对应的匹配关系。
图8示出了本发明提供的高分辨率图像匹配系统的结构框图，为了便于说明，图中仅给出了与本发明实施例相关的部分。
高分辨率图像匹配系统包括：
下采样模块11，用于对初始高分辨率图像进行区域保真下采样，得到多层次的低分辨率图像；
局部匹配模块12，用于采用带有全局探针的方式对得到的多层次的所述低分辨率图像进行局部匹配，得到低分辨率图像的匹配结果；
反向求精模块13，用于利用图像匹配的整体一致性，对得到的所述低分辨率图像的匹配结果进行反向求精，得到各级高分辨率图像的匹配结果，直至得到初始分辨率图像的匹配结果。
如图9所示，下采样模块11具体包括：
值初始化模块14，用于初始化像素个数R的值，所述像素个数R作为停止采样条件；
像素值确认模块15，用于将得到像素个数R的值作为下一级分辨率图像对应位置的像素值；
最低分辨率图像获取模块16，用于依次执行上述下采样过程，直到得到的图像分辨率满足采样停止条件，停止采样过程，最后得到的图像即为最低分辨率的图像。
如图10所示，所述局部匹配模块12具体包括：
探针初始化模块17，用于初始化全局探针数量、方向和长度；
像素匹配模块18，用于对得到的所述最低分辨率图像进行逐像素匹配，计算得到各个匹配结果对应的匹配代价；
大小关系计算模块19，用于根据确定的探针数量、方向和长度，计算 所述最低分辨率图像在探针上各个位置处的像素值和中心像素值的大小关系，以二进制值表示；
探针距离计算模块20，用于对所有待匹配像素，计算其对应的探针之间的距离；
最终匹配代价计算模块21，用于对每个待匹配像素，依据计算得到的各个匹配结果对应的匹配代价和得到的对应的探针之间距离进行加权求和，得到的加权求和结果作为各个待匹配像素的最终匹配代价；
最终匹配结果确认模块22，用于选取匹配代价最小的候选像素作为所述低分辨率图像的最终匹配结果。
如图11所示，所述反向求精模块13具体包括：
配对处理模块23，用于在得到的低分辨率图像的匹配结果中，对每个匹配像素执行配对处理，找到其各自在上一级分辨率图像中对应的像素区域；
最优像素匹配关系计算模块24，用于对得到的上一级分辨率图像中对应的像素区域，进行区域内像素的匹配计算，得到像素区域内最优像素匹配关系；
匹配关系确定模块25，用于将得到的所述像素区域内最优像素匹配关系确定为当前分辨率图像之间的匹配关系；
循环执行模块26，用于循环执行像素区域内最优像素匹配关系的匹配计算和当前分辨率图像之间的匹配关系确定操作步骤，直到得到最高分辨率图像的匹配结果，作为最终的初始分辨率图像匹配结果。
上述各个模块的功能如上述方法实施例所记载，在此不再赘述。
在本发明实施例中，对初始高分辨率图像进行区域保真下采样，得到多层次的低分辨率图像；采用带有全局探针的方式对得到的多层次的所述低分辨率图像进行局部匹配，得到低分辨率图像的匹配结果；利用图像匹配的整体一致性，对得到的所述低分辨率图像的匹配结果进行反向求精，得到各级高分辨率图像的匹配结果，直至得到初始分辨率图像的匹配结果，从而既能降低图像匹配过程的计算量，又能提高匹配结果的精度，再基于整体一致性，通过反向求精得到高分辨率图像的匹配结果。
以上各实施例仅用以说明本发明的技术方案，而非对其限制；尽管参照前述各实施例对本发明进行了详细的说明，本领域的普通技术人员应当理解：其依然可以对前述各实施例所记载的技术方案进行修改，或者对其中部分或者全部技术特征进行等同替换；而这些修改或者替换，并不使相应技术方案的本质脱离本发明各实施例技术方案的范围，其均应涵盖在本发明的权利要求和说明书的范围当中。
Claims (10)
- 一种高分辨率图像匹配方法，其特征在于，所述方法包括下述步骤：对初始高分辨率图像进行区域保真下采样，得到多层次的低分辨率图像；采用带有全局探针的方式对得到的多层次的所述低分辨率图像进行局部匹配，得到低分辨率图像的匹配结果；利用图像匹配的整体一致性，对得到的所述低分辨率图像的匹配结果进行反向求精，得到各级高分辨率图像的匹配结果，直至得到初始分辨率图像的匹配结果。
- 根据权利要求1所述的高分辨率图像匹配方法，其特征在于，所述对初始高分辨率图像进行区域保真下采样，得到多层次的低分辨率图像的步骤具体包括下述步骤：初始化像素个数R的值，所述像素个数R作为停止采样条件；将得到像素个数R的值作为下一级分辨率图像对应位置的像素值；依次执行上述下采样过程，直到得到的图像分辨率满足采样停止条件，停止采样过程，最后得到的图像即为最低分辨率的图像。
- 根据权利要求2所述的高分辨率图像匹配方法，其特征在于，在初始高分辨率图像过程中，综合上下左右相邻的四个像素信息，得到一个值，得到的每个值包含了相邻四个图像的信息。
- 根据权利要求2所述的高分辨率图像匹配方法，其特征在于，所述采用带有全局探针的方式对得到的多层次的所述低分辨率图像进行局部匹配，得到低分辨率图像的匹配结果的步骤具体包括下述步骤：初始化全局探针数量、方向和长度；对得到的所述最低分辨率图像进行逐像素匹配，计算得到各个匹配结果对应的匹配代价；根据确定的探针数量、方向和长度，计算所述最低分辨率图像在探针上各个位置处的像素值和中心像素值的大小关系，以二进制值表示；对所有待匹配像素，计算其对应的探针之间的距离；对每个待匹配像素，依据计算得到的各个匹配结果对应的匹配代价和 得到的对应的探针之间距离进行加权求和，得到的加权求和结果作为各个待匹配像素的最终匹配代价；选取匹配代价最小的候选像素作为所述低分辨率图像的最终匹配结果。
- 根据权利要求1所述的高分辨率图像匹配方法，其特征在于，所述利用图像匹配的整体一致性，对得到的所述低分辨率图像的匹配结果进行反向求精，得到各级高分辨率图像的匹配结果，直至得到初始分辨率图像的匹配结果的步骤具体包括下述步骤：在得到的低分辨率图像的匹配结果中，对每个匹配像素执行配对处理，找到其各自在上一级分辨率图像中对应的像素区域；对得到的上一级分辨率图像中对应的像素区域，进行区域内像素的匹配计算，得到像素区域内最优像素匹配关系；将得到的所述像素区域内最优像素匹配关系确定为当前分辨率图像之间的匹配关系；循环执行像素区域内最优像素匹配关系的匹配计算和当前分辨率图像之间的匹配关系确定操作步骤，直到得到最高分辨率图像的匹配结果，作为最终的初始分辨率图像匹配结果。
- 一种高分辨率图像匹配系统，其特征在于，所述系统包括：下采样模块，用于对初始高分辨率图像进行区域保真下采样，得到多层次的低分辨率图像；局部匹配模块，用于采用带有全局探针的方式对得到的多层次的所述低分辨率图像进行局部匹配，得到低分辨率图像的匹配结果；反向求精模块，用于利用图像匹配的整体一致性，对得到的所述低分辨率图像的匹配结果进行反向求精，得到各级高分辨率图像的匹配结果，直至得到初始分辨率图像的匹配结果。
- 根据权利要求1所述的高分辨率图像匹配系统，其特征在于，所述下采样模块具体包括：值初始化模块，用于初始化像素个数R的值，所述像素个数R作为停止采样条件；像素值确认模块，用于将得到像素个数R的值作为下一级分辨率图像对应位置的像素值；最低分辨率图像获取模块，用于依次执行上述下采样过程，直到得到的图像分辨率满足采样停止条件，停止采样过程，最后得到的图像即为最低分辨率的图像。
- 根据权利要求7所述的高分辨率图像匹配系统，其特征在于，在初始高分辨率图像过程中，综合上下左右相邻的四个像素信息，得到一个值，得到的每个值包含了相邻四个图像的信息。
- 根据权利要求7所述的高分辨率图像匹配系统，其特征在于，所述局部匹配模块具体包括：探针初始化模块，用于初始化全局探针数量、方向和长度；像素匹配模块，用于对得到的所述最低分辨率图像进行逐像素匹配，计算得到各个匹配结果对应的匹配代价；大小关系计算模块，用于根据确定的探针数量、方向和长度，计算所述最低分辨率图像在探针上各个位置处的像素值和中心像素值的大小关系，以二进制值表示；探针距离计算模块，用于对所有待匹配像素，计算其对应的探针之间的距离；最终匹配代价计算模块，用于对每个待匹配像素，依据计算得到的各个匹配结果对应的匹配代价和得到的对应的探针之间距离进行加权求和，得到的加权求和结果作为各个待匹配像素的最终匹配代价；最终匹配结果确认模块，用于选取匹配代价最小的候选像素作为所述低分辨率图像的最终匹配结果。
- 根据权利要求6所述的高分辨率图像匹配系统，其特征在于，所述反向求精模块具体包括：配对处理模块，用于在得到的低分辨率图像的匹配结果中，对每个匹配像素执行配对处理，找到其各自在上一级分辨率图像中对应的像素区域；最优像素匹配关系计算模块，用于对得到的上一级分辨率图像中对应的像素区域，进行区域内像素的匹配计算，得到像素区域内最优像素匹配 关系；匹配关系确定模块，用于将得到的所述像素区域内最优像素匹配关系确定为当前分辨率图像之间的匹配关系；循环执行模块，用于循环执行像素区域内最优像素匹配关系的匹配计算和当前分辨率图像之间的匹配关系确定操作步骤，直到得到最高分辨率图像的匹配结果，作为最终的初始分辨率图像匹配结果。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| US17/614,850 US12062223B2 (en) | 2019-06-29 | 2019-10-31 | High-resolution image matching method and system |
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910580849.3 | 2019-06-29 | ||
| CN201910580849.3A CN110363235B (zh) | 2019-06-29 | 2019-06-29 | 一种高分辨率图像匹配方法及系统 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2021000471A1 true WO2021000471A1 (zh) | 2021-01-07 |
Family
ID=68216725
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/CN2019/114525 WO2021000471A1 (zh) | 2019-06-29 | 2019-10-31 | 一种高分辨率图像匹配方法及系统 |
Country Status (3)
| Country | Link |
|---|---|
| US (1) | US12062223B2 (zh) |
| CN (1) | CN110363235B (zh) |
| WO (1) | WO2021000471A1 (zh) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110363235B (zh) | 2019-06-29 | 2021-08-06 | 苏州浪潮智能科技有限公司 | 一种高分辨率图像匹配方法及系统 |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100134496A1 (en) * | 2008-12-01 | 2010-06-03 | Vasudev Bhaskaran | Bit resolution enhancement |
| CN102629374A (zh) * | 2012-02-29 | 2012-08-08 | 西南交通大学 | 基于子空间投影和邻域嵌入的图像超分辨率重建方法 |
| US20150003725A1 (en) * | 2013-06-28 | 2015-01-01 | Canon Kabushiki Kaisha | Depth constrained superpixel-based depth map refinement |
| CN108492326A (zh) * | 2018-01-31 | 2018-09-04 | 北京大学深圳研究生院 | 分辨率从低到高的逐渐精细化的立体匹配方法及系统 |
| CN110363235A (zh) * | 2019-06-29 | 2019-10-22 | 苏州浪潮智能科技有限公司 | 一种高分辨率图像匹配方法及系统 |
Family Cites Families (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8339470B2 (en) * | 2009-11-30 | 2012-12-25 | Indian Institute Of Technology Madras | Method and system for generating a high resolution image |
| CN105335952B (zh) * | 2014-06-19 | 2018-04-17 | 株式会社理光 | 匹配代价计算方法和装置、以及视差值计算方法和设备 |
| CN104156957B (zh) * | 2014-08-06 | 2017-02-01 | 昆山天工智能科技有限公司 | 一种稳定高效的高分辨率立体匹配方法 |
| CN104184981B (zh) * | 2014-08-27 | 2017-12-15 | 深圳市华星光电技术有限公司 | 一种基于缩减像素采样处理的低解析度显示方法及设备 |
| CN105488776B (zh) * | 2014-10-10 | 2018-05-08 | 北京大学 | 超分辨率图像重建方法和装置 |
| US9767533B2 (en) * | 2015-02-16 | 2017-09-19 | Adobe Systems Incorporated | Image resolution enhancement based on data from related images |
| TWI744245B (zh) * | 2015-08-19 | 2021-11-01 | 新加坡商海特根微光學公司 | 產生具有減少過度平滑之視差圖 |
| CN105466399B (zh) * | 2016-01-11 | 2019-09-06 | 中测新图(北京)遥感技术有限责任公司 | 快速半全局密集匹配方法和装置 |
| CN108010082B (zh) * | 2017-12-28 | 2021-07-23 | 上海觉感视觉科技有限公司 | 一种几何匹配的方法 |
| CN108460795A (zh) * | 2018-03-05 | 2018-08-28 | 成都通甲优博科技有限责任公司 | 倒金字塔式双目立体匹配方法及系统 |
| CN108682029A (zh) * | 2018-03-22 | 2018-10-19 | 深圳飞马机器人科技有限公司 | 多尺度密集匹配方法和系统 |
| CN109711457A (zh) * | 2018-12-20 | 2019-05-03 | 江南大学 | 一种基于改进hu不变矩的快速图像匹配方法及其应用 |
-
2019
- 2019-06-29 CN CN201910580849.3A patent/CN110363235B/zh active Active
- 2019-10-31 US US17/614,850 patent/US12062223B2/en active Active
- 2019-10-31 WO PCT/CN2019/114525 patent/WO2021000471A1/zh active Application Filing
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100134496A1 (en) * | 2008-12-01 | 2010-06-03 | Vasudev Bhaskaran | Bit resolution enhancement |
| CN102629374A (zh) * | 2012-02-29 | 2012-08-08 | 西南交通大学 | 基于子空间投影和邻域嵌入的图像超分辨率重建方法 |
| US20150003725A1 (en) * | 2013-06-28 | 2015-01-01 | Canon Kabushiki Kaisha | Depth constrained superpixel-based depth map refinement |
| CN108492326A (zh) * | 2018-01-31 | 2018-09-04 | 北京大学深圳研究生院 | 分辨率从低到高的逐渐精细化的立体匹配方法及系统 |
| CN110363235A (zh) * | 2019-06-29 | 2019-10-22 | 苏州浪潮智能科技有限公司 | 一种高分辨率图像匹配方法及系统 |
Also Published As
| Publication number | Publication date |
|---|---|
| US12062223B2 (en) | 2024-08-13 |
| CN110363235B (zh) | 2021-08-06 |
| CN110363235A (zh) | 2019-10-22 |
| US20220230412A1 (en) | 2022-07-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111862126B (zh) | 深度学习与几何算法结合的非合作目标相对位姿估计方法 | |
| CN110298298B (zh) | 目标检测及目标检测网络的训练方法、装置及设备 | |
| CN109377530B (zh) | 一种基于深度神经网络的双目深度估计方法 | |
| CN110427968B (zh) | 一种基于细节增强的双目立体匹配方法 | |
| CN109712071B (zh) | 基于航迹约束的无人机图像拼接与定位方法 | |
| CN113657388B (zh) | 一种融合图像超分辨率重建的图像语义分割方法 | |
| CN110633661A (zh) | 一种融合语义分割的遥感图像目标检测方法 | |
| CN108364272A (zh) | 一种高性能红外-可见光融合探测方法 | |
| CN111861880B (zh) | 基于区域信息增强与块自注意力的图像超分与融合方法 | |
| CN110910437B (zh) | 一种复杂室内场景的深度预测方法 | |
| CN113221925B (zh) | 一种基于多尺度图像的目标检测方法及装置 | |
| CN111105452A (zh) | 基于双目视觉的高低分辨率融合立体匹配方法 | |
| CN115035172B (zh) | 基于置信度分级及级间融合增强的深度估计方法及系统 | |
| CN116129037A (zh) | 视触觉传感器及其三维重建方法、系统、设备及存储介质 | |
| CN109741358A (zh) | 基于自适应超图学习的超像素分割方法 | |
| Krauss et al. | Deterministic guided lidar depth map completion | |
| CN111739071A (zh) | 基于初始值的快速迭代配准方法、介质、终端和装置 | |
| WO2021000471A1 (zh) | 一种高分辨率图像匹配方法及系统 | |
| Sun et al. | Decoupled feature pyramid learning for multi-scale object detection in low-altitude remote sensing images | |
| CN117593620A (zh) | 一种基于相机和激光雷达融合的多目标检测方法及装置 | |
| WO2022188102A1 (zh) | 深度图像的修复方法及装置、摄像头组件及电子设备 | |
| CN116503686B (zh) | 图像矫正模型的训练方法、图像矫正方法、装置及介质 | |
| EP4187483A1 (en) | Apparatus and method with image processing | |
| CN115619678A (zh) | 一种图像变形的矫正方法、装置、计算机设备及存储介质 | |
| JP2009146150A (ja) | 特徴位置検出方法及び特徴位置検出装置 |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 19935936 Country of ref document: EP Kind code of ref document: A1 |
|
| NENP | Non-entry into the national phase |
Ref country code: DE |
|
| 122 | Ep: pct application non-entry in european phase |
Ref document number: 19935936 Country of ref document: EP Kind code of ref document: A1 |