WO2022073249A1 - 一种基于双向最优匹配点对的图像匹配方法 - Google Patents
一种基于双向最优匹配点对的图像匹配方法 Download PDFInfo
- Publication number
- WO2022073249A1 WO2022073249A1 PCT/CN2020/120449 CN2020120449W WO2022073249A1 WO 2022073249 A1 WO2022073249 A1 WO 2022073249A1 CN 2020120449 W CN2020120449 W CN 2020120449W WO 2022073249 A1 WO2022073249 A1 WO 2022073249A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- matching point
- optimal matching
- point pairs
- optimal
- bidirectional
- Prior art date
Links
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
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
- G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
-
- 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
Definitions
- the invention belongs to the technical field of image processing, and in particular relates to an image matching method based on two-way optimal matching point pairs.
- the general process of image matching based on feature points is: extract the features of the image, generate feature descriptors, find several pairs of feature points that match each other through the pairwise comparison of the features of the two images (feature points + descriptors), and eliminate the wrong matching points. Yes, calculate the transformation relationship between the two graphs according to the reserved matching point pairs.
- the random sample consensus algorithm can estimate the parameters of the mathematical model from the observed data containing abnormal data, eliminate invalid samples, and retain valid samples.
- Step1 randomly select a subset from the set of matching point pairs
- Step2 uses the matching point pairs in the subset to calculate the transformation matrix, such as the homography matrix
- Step3 uses the transformation matrix to calculate whether each matching point pair in the matching point pair set conforms to the transformation matrix, and the matching point pair that conforms to the transformation is marked as an inner point, otherwise it is marked as an outer point;
- Step 4 Calculate the error probability of the loop according to the number of inner and outer points, or calculate the mapping error of the inner point according to the transformation relationship. If the preset termination condition is not met, return to Step 1 to continue the calculation until the requirements are met.
- Exhaustive method For each feature point in the target image, the exhaustive method is used to traverse each feature point in the original image, and the point with the highest similarity is taken as the matching point of the target image feature point in the original image, thereby obtaining a set of matching point pairs.
- the ratio method the method proposed by SIFT author Lowe. Take a key point in the target image and find the first two key points with the closest Euclidean distance to the original image. Among these two key points, if the closest distance divided by the second closest distance is less than a certain ratio threshold, the pair of matching points is accepted.
- the above-mentioned first scheme has a low matching accuracy rate, and the obtained set of matching point pairs will contain a large number of incorrect matching points, thereby greatly increasing the processing complexity and calculation amount when eliminating incorrect matching point pairs.
- the second scheme above depends on the selection of the threshold.
- the general principle is that when the threshold is reduced, the matching correct rate increases, but the number of matching point pairs decreases; when the threshold is increased, the matching correct rate decreases, including a large number of incorrect matching point pairs.
- how to select an appropriate threshold is a difficult problem to solve.
- the similarity of local areas is relatively large, and the next closest distance is often close to the correct matching point pair. The selection of the threshold cannot solve the matching problem of such images.
- the invention provides an image matching method based on bidirectional optimal matching point pairs, and proposes the concept of bidirectional optimal matching point pairs in the feature matching stage, so that the number of incorrect matching point pairs contained in the reserved matching point pairs is as small as possible, The number of correct matching point pairs is as large as possible, thereby reducing the computational complexity of eliminating incorrect matching point pairs, and quickly calculating a stable transformation relationship between two images.
- the present invention provides an image matching method based on two-way optimal matching point pairs, the specific steps are as follows, and it is characterized in that:
- Step S1 use the exhaustive method to perform bidirectional matching to obtain the optimal matching points of the feature points in the original image in the target image, and the optimal matching points of the feature points in the target image in the original image, forming two optimal matching point pairs. gather;
- Step S2 according to the above-mentioned two sets of optimal matching point pairs, calculate a set of two-way optimal matching point pairs;
- Step S3 using the RANSAC algorithm to eliminate the erroneous matching points in the set of bidirectional optimal matching point pairs;
- Step S4 according to the reserved bidirectional optimal matching point pair, calculate the transformation relationship of the two images
- Step S5 select a matching point pair that satisfies the above-mentioned transformation relationship from the two sets of optimal matching point pairs obtained in step S1;
- step S6 the retained bidirectional optimal matching point pair and the optimal matching point pair satisfying the transformation relationship are the acceptable correct matching point pair between the two images.
- step S1 the specific steps of step S1 are as follows:
- step S2 the specific steps of step S2 are as follows:
- the two-way optimal matching point pair is defined as:
- the similarity is calculated by using the cosine distance or the Euclidean distance, and the similarity can be obtained by calculating the above two methods.
- the present invention provides an image matching method based on two-way optimal matching point pairs, which has the following advantages:
- RANSAC Use the RANSAC algorithm to eliminate the wrong matching points in the set of bidirectional optimal matching point pairs. Since RANSAC is a random sampling method, its optimization process is based on loop iteration, and the number of iterations is uncontrollable, and the optimization results cannot be guaranteed without full-space search. Therefore, if the valid data in the matching point pair is the majority and the invalid data is very small, it is more efficient to determine the parameters and errors of the transformation model by the least squares method or similar methods. Fast and stable method. In the set of bidirectional optimal matching point pairs described in the present application, the proportion of incorrect matching point pairs is small, and a sufficiently accurate and stable transformation relationship can often be obtained by the least squares method.
- the present invention can keep the correct number of matching point pairs as much as possible, greatly reducing the computational complexity when correcting the matching point pairs. For the relabeling method, ensure that the two images have enough matching point pairs.
- Figure 1 is a flow chart of the method of the present invention.
- the invention provides an image matching method based on bidirectional optimal matching point pairs, and proposes the concept of bidirectional optimal matching point pairs in the feature matching stage, so that the number of incorrect matching point pairs contained in the reserved matching point pairs is as small as possible, The number of correct matching point pairs is as large as possible, thereby reducing the computational complexity of eliminating incorrect matching point pairs, and quickly calculating a stable transformation relationship between two images.
- the present invention provides an image matching method based on bidirectional optimal matching point pairs as shown in FIG. 1 , and the specific steps are as follows:
- Step S1 use the exhaustive method to perform two-way matching to obtain the optimal matching point of the feature points in the original image in the target image, and the optimal matching point of the feature points in the target image in the original image, forming two optimal matching point pairs. gather;
- step S1 The specific steps of step S1 are as follows:
- Step S2 according to the above-mentioned two sets of optimal matching point pairs, calculate a set of two-way optimal matching point pairs;
- step S2 The specific steps of step S2 are as follows:
- the two-way optimal matching point pair is defined as:
- Step S3 using the RANSAC algorithm to eliminate the erroneous matching points in the set of bidirectional optimal matching point pairs;
- Step S4 according to the reserved bidirectional optimal matching point pair, calculate the transformation relationship of the two images
- Step S5 select a matching point pair that satisfies the above-mentioned transformation relationship from the two sets of optimal matching point pairs obtained in step S1;
- step S6 the retained bidirectional optimal matching point pair and the optimal matching point pair satisfying the transformation relationship are the acceptable correct matching point pair between the two images.
- RANSAC is a random sampling method. Its optimization process is based on loop iterations, and the number of iterations is uncontrollable. The optimization results cannot be guaranteed to be consistent without full-space search. When the data is in the majority and the invalid data is small, we can determine the parameters and errors of the transformation model by least squares or similar methods.
- the proportion of wrong matching point pairs is very small, and a sufficiently accurate and stable transformation relationship can often be obtained by the least square method.
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Theoretical Computer Science (AREA)
- Multimedia (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Databases & Information Systems (AREA)
- Computing Systems (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
Abstract
一种基于双向最优匹配点对的图像匹配方法,S1、利用穷举法进行双向匹配,得到原图像中特征点在目标图像中的最优匹配点,和目标图像中特征点在原图像中的最优匹配点,组成两个最优匹配点对集合;S2、根据上述两个最优匹配点对集合,计算双向最优匹配点对集合;S3、剔除双向最优匹配点对集合中的错误匹配点;S4、根据保留的双向最优匹配点对,计算两幅图像的变换关系;S5、从步骤S1中所得两组最优匹配点对集合中挑选满足上述变换关系的匹配点对;S6、得到两幅图像之间可接受的正确匹配点对。该方法能在只引入少量错误匹配点对的前提下保留足够多正确的匹配点对数,大大减少剔除错误匹配点对时的计算量,并计算出两幅图像之间精准稳定的变换关系。
Description
本发明属于图像处理技术领域,特别是涉及到一种基于双向最优匹配点对的图像匹配方法。
基于特征点实现图像匹配的一般过程是:提取图像的特征,生成特征描述子,通过两图特征(特征点+描述子)的两两比较找出相互匹配的若干对特征点,剔除错误匹配点对,根据保留的匹配点对计算两图的变换关系。
匹配点对校正时,往往采用RANSAC算法剔除错误匹配点对。随机抽样一致算法(random sample consensus,RANSAC),可从包含异常数据的被观测数据中估算出数学模型的参数,剔除无效样本,保留有效样本。
RANSAC算法的处理步骤如下:
Step1从匹配点对集合中随机抽选一个子集;
Step2利用子集中的匹配点对计算出变换矩阵,如单应矩阵;
Step3利用变换矩阵,计算匹配点对集合中的每个匹配点对是否符合变换矩阵,符合变换的匹配点对标记为内点,否则标记为外点;
Step4根据内外点的个数计算该循环的错误概率,或根据变换关系计算内点的映射误差,若不满足预设的终止条件,则返回 Step1继续计算,直到满足要求为止。
通过RANSAC算法的迭代过程,剔除错误匹配点对,并根据保留的匹配点对计算变换关系。
目前,在特征匹配阶段获取匹配点对,普遍的做法有以下两种:
穷举法。对于目标图像中的每一个特征点,利用穷举法遍历原图像中的每个特征点,取相似度最高的点作为目标图像特征点在原图像中的匹配点,从而得到一组匹配点对。
比值法,即SIFT作者lowe提出的方法。取目标图像中的某个关键点,找出其与原图像中欧式距离最近的前两个关键点,在这两个关键点中,如果最近的距离除以次近的距离少于某个比例阈值,则接受这一对匹配点。
上述第1种方案匹配正确率低,得到的匹配点对集合中将包含大量的错误匹配点,从而大大增加剔除错误匹配点对时的处理复杂度和计算量。
上述第2种方案依赖于阈值的选取,一般原则是,减小阈值则匹配正确率上升,但匹配点对数目减少;增加阈值则匹配正确率下降,包含大量错误匹配点对。针对不同质量的图像,如何选取合适的阈值,是很难解决的问题。另外,针对指纹图像,局部区域相似性较大,次近的距离往往接近于正确匹配点对,阈值的选取无法解决这类图像的匹配问题。
因此,如何保留特征匹配点对,减少匹配点校正时的搜索空间, 降低计算复杂度是特征匹配的重要难题。
发明内容
为了解决上述存在问题。本发明提供一种基于双向最优匹配点对的图像匹配方法,在特征匹配阶段提出双向最优匹配点对的概念,使保留的匹配点对中包含的错误匹配点对数尽可能的少,正确匹配点对数尽可能多,从而降低剔除错误匹配点对的计算复杂度,快速计算出两幅图像之间稳定的变换关系。
本发明提供一种基于双向最优匹配点对的图像匹配方法,具体步骤如下,其特征在于:
步骤S1、利用穷举法进行双向匹配,得到原图像中特征点在目标图像中的最优匹配点,和目标图像中特征点在原图像中的最优匹配点,组成两个最优匹配点对集合;
步骤S2、根据上述两个最优匹配点对集合,计算双向最优匹配点对集合;
步骤S3、利用RANSAC算法剔除双向最优匹配点对集合中的错误匹配点;
步骤S4、根据保留的双向最优匹配点对,计算两幅图像的变换关系;
步骤S5、从步骤S1中所得两组最优匹配点对集合中挑选满足上述变换关系的匹配点对;
步骤S6、保留的双向最优匹配点对,和满足变换关系的最优匹配点对,即为两幅图像之间可接受的正确匹配点对。
作为本发明进一步细化,步骤S1具体步骤如下:
记原图像中提取的特征点集为T:T={t
i},i=1,2,...,m
记目标图像中提取的特征点集为V:V={v
j},j=1,2,...,n
记原图像中特征点t
i与目标图像中特征点v
j的相似度为s
ij;
则定义最优匹配点对为:
作为本发明进一步细化,步骤S2具体步骤如下:
作为本发明进一步细化,所述相似度采用余弦距离或欧氏距离进行计算,相似度可以采用以上两种方式进行计算获得。
本发明提供一种基于双向最优匹配点对的图像匹配方法,具有如下优点:
1)与现有的技术相比,穷举法得到的最优匹配点对集中存在一对多的匹配点对,这是明显的错误匹配点对。本发明定义的双向最优匹配点对中只存在一对一的匹配点对,错误匹配点对数明显降低,且避免了比值法中对阈值选取的依赖;
2)利用RANSAC算法剔除双向最优匹配点对集合中的错误匹配 点,由于RANSAC是一种随机抽样方法,其优化过程基于循环迭代,迭代次数不可控,不进行全空间搜索时优化结果不能保证一致,而全空间搜索时计算量过大,因此,若匹配点对中有效数据占大多数、无效数据很少时,通过最小二乘法或类似的方法来确定变换模型的参数和误差,是更快捷稳定的方法。本申请所述双向最优匹配点对集合中,错误匹配点对的占比很小,通过最小二乘法往往即可得到足够精准稳定的变换关系。
3)本申请在特征匹配阶段,本发明能尽可能保留正确的匹配点对数,大大减少匹配点对校正时的计算复杂度,在计算出稳定变换关系后,通过将非双向最优匹配点对重新标记的方式,保证两幅图像具有足够多的匹配点对数。
图1为本发明方法的流程图。
下面结合附图与具体实施方式对本发明作进一步详细描述:
本发明提供一种基于双向最优匹配点对的图像匹配方法,在特征匹配阶段提出双向最优匹配点对的概念,使保留的匹配点对中包含的错误匹配点对数尽可能的少,正确匹配点对数尽可能多,从而降低剔除错误匹配点对的计算复杂度,快速计算出两幅图像之间稳定的变换关系。
作为发明的一种实施例,本发明提供如图1所示一种基于双向最优匹配点对的图像匹配方法,具体步骤如下:
步骤S1、利用穷举法进行双向匹配,得到原图像中特征点在目标图像中的最优匹配点,和目标图像中特征点在原图像中的最优匹配点,组成两个最优匹配点对集合;
步骤S1具体步骤如下:
记原图像中提取的特征点集为T:T={t
i},i=1,2,...,m
记目标图像中提取的特征点集为V:V={v
j},j=1,2,...,n
记原图像中特征点t
i与目标图像中特征点v
j的相似度为s
ij;
则定义最优匹配点对为:
上述
和
均可称为最优匹配点对,其含义分别为:
表示v
y为t
x在目标图像中相似度最大的匹配点,
表示t
x为v
y在原图像中相似度最大的匹配点,所述相似度采用余弦距离或欧氏距离进行计算,相似度可以采用以上两种方式进行计算获得;
步骤S2、根据上述两个最优匹配点对集合,计算双向最优匹配点对集合;
步骤S2具体步骤如下:
步骤S3、利用RANSAC算法剔除双向最优匹配点对集合中的错误匹配点;
步骤S4、根据保留的双向最优匹配点对,计算两幅图像的变换关系;
步骤S5、从步骤S1中所得两组最优匹配点对集合中挑选满足上述变换关系的匹配点对;
步骤S6、保留的双向最优匹配点对,和满足变换关系的最优匹配点对,即为两幅图像之间可接受的正确匹配点对。
RANSAC是一种随机抽样方法,其优化过程基于循环迭代,迭代次数不可控,不进行全空间搜索时优化结果不能保证一致,而全空间搜索时计算量过大,因此,若匹配点对中有效数据占大多数、无效数据很少时,我们可以通过最小二乘法或类似的方法来确定变换模型的参数和误差。
本发明所述双向最优匹配点对集合中,错误匹配点对的占比很小,通过最小二乘法往往即可得到足够精准稳定的变换关系。
以上所述,仅是本发明的较佳实施例而已,并非是对本发明作任何其他形式的限制,而依据本发明的技术实质所作的任何修改或等同变化,仍属于本发明所要求保护的范围。
Claims (4)
- 一种基于双向最优匹配点对的图像匹配方法,具体步骤如下,其特征在于:步骤S1、利用穷举法进行双向匹配,得到原图像中特征点在目标图像中的最优匹配点,和目标图像中特征点在原图像中的最优匹配点,组成两个最优匹配点对集合;步骤S2、根据上述两个最优匹配点对集合,计算双向最优匹配点对集合;步骤S3、利用RANSAC算法剔除双向最优匹配点对集合中的错误匹配点;步骤S4、根据保留的双向最优匹配点对,计算两幅图像的变换关系;步骤S5、从步骤S1中所得两组最优匹配点对集合中挑选满足上述变换关系的匹配点对;步骤S6、保留的双向最优匹配点对,和满足变换关系的最优匹配点对,即为两幅图像之间可接受的正确匹配点对。
- 根据权利要求2所述的一种基于双向最优匹配点对的图像匹配方法,其特征在于:所述相似度采用余弦距离或欧氏距离进行计算。
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011079952.9A CN112364879A (zh) | 2020-10-10 | 2020-10-10 | 一种基于双向最优匹配点对的图像匹配方法 |
CN202011079952.9 | 2020-10-10 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2022073249A1 true WO2022073249A1 (zh) | 2022-04-14 |
Family
ID=74507624
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/CN2020/120449 WO2022073249A1 (zh) | 2020-10-10 | 2020-10-12 | 一种基于双向最优匹配点对的图像匹配方法 |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN112364879A (zh) |
WO (1) | WO2022073249A1 (zh) |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20120148164A1 (en) * | 2010-12-08 | 2012-06-14 | Electronics And Telecommunications Research Institute | Image matching devices and image matching methods thereof |
CN103400388A (zh) * | 2013-08-06 | 2013-11-20 | 中国科学院光电技术研究所 | 一种利用RANSAC消除Brisk关键点错误匹配点对的方法 |
CN110009670A (zh) * | 2019-03-28 | 2019-07-12 | 上海交通大学 | 基于fast特征提取和piifd特征描述的异源图像配准方法 |
CN111104922A (zh) * | 2019-12-30 | 2020-05-05 | 深圳纹通科技有限公司 | 一种基于有序抽样的特征匹配算法 |
CN111144338A (zh) * | 2019-12-30 | 2020-05-12 | 深圳纹通科技有限公司 | 一种基于特征点拓扑结构的特征匹配算法 |
Family Cites Families (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103116891B (zh) * | 2013-03-12 | 2015-08-12 | 上海海事大学 | 一种基于双向邻域过滤策略的遥感图像配准方法 |
US9400939B2 (en) * | 2014-04-13 | 2016-07-26 | International Business Machines Corporation | System and method for relating corresponding points in images with different viewing angles |
CN104596519B (zh) * | 2015-02-17 | 2017-06-13 | 哈尔滨工业大学 | 基于ransac算法的视觉定位方法 |
-
2020
- 2020-10-10 CN CN202011079952.9A patent/CN112364879A/zh active Pending
- 2020-10-12 WO PCT/CN2020/120449 patent/WO2022073249A1/zh active Application Filing
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20120148164A1 (en) * | 2010-12-08 | 2012-06-14 | Electronics And Telecommunications Research Institute | Image matching devices and image matching methods thereof |
CN103400388A (zh) * | 2013-08-06 | 2013-11-20 | 中国科学院光电技术研究所 | 一种利用RANSAC消除Brisk关键点错误匹配点对的方法 |
CN110009670A (zh) * | 2019-03-28 | 2019-07-12 | 上海交通大学 | 基于fast特征提取和piifd特征描述的异源图像配准方法 |
CN111104922A (zh) * | 2019-12-30 | 2020-05-05 | 深圳纹通科技有限公司 | 一种基于有序抽样的特征匹配算法 |
CN111144338A (zh) * | 2019-12-30 | 2020-05-12 | 深圳纹通科技有限公司 | 一种基于特征点拓扑结构的特征匹配算法 |
Also Published As
Publication number | Publication date |
---|---|
CN112364879A (zh) | 2021-02-12 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN113012212B (zh) | 一种基于深度信息融合的室内场景三维点云重建方法和*** | |
CN108764041B (zh) | 用于下部遮挡人脸图像的人脸识别方法 | |
She | An iterative algorithm for fitting nonconvex penalized generalized linear models with grouped predictors | |
WO2020119619A1 (zh) | 一种基于3d目标分类和场景语义分割的网络优化结构 | |
CN109697692B (zh) | 一种基于局部结构相似的特征匹配方法 | |
CN111815686B (zh) | 基于几何特征由粗到细点云配准方法 | |
WO2019205813A1 (zh) | 移动机器人的陀螺仪数据校正方法、设备和存储介质 | |
WO2021072891A1 (zh) | 知识图谱的关系对齐方法、装置、设备及存储介质 | |
WO2023125456A1 (zh) | 基于多层次变分自动编码器的高光谱图像特征提取方法 | |
CN113095385B (zh) | 一种基于全局和局部特征描述的多模图像匹配方法 | |
WO2022073249A1 (zh) | 一种基于双向最优匹配点对的图像匹配方法 | |
CN114862650A (zh) | 一种神经网络水印嵌入方法以及验证方法 | |
WO2021223505A1 (zh) | 一种基于广义Pascal映射的椭圆检测加速方法 | |
CN114494368A (zh) | 一种结合降维投影和特征匹配的低重叠率点云配准方法 | |
Guo et al. | Triplet relationship guided sampling consensus for robust model estimation | |
CN111881989B (zh) | 一种高光谱图像分类方法 | |
CN111144469B (zh) | 基于多维关联时序分类神经网络的端到端多序列文本识别方法 | |
CN110400300B (zh) | 基于块匹配自适应权重稀疏表示的病变血管精确检测方法 | |
CN113470085B (zh) | 一种基于改进的ransac的图像配准方法 | |
CN111104922A (zh) | 一种基于有序抽样的特征匹配算法 | |
Wei et al. | L2-norm prototypical networks for tackling the data shift problem in scene classification | |
CN111192302A (zh) | 一种基于运动平滑性和ransac算法的特征匹配方法 | |
Zhang et al. | Research on finger vein recognition based on sub-convolutional neural network | |
CN112837275B (zh) | 胶囊内窥镜图像器官分类方法、装置、设备及存储介质 | |
CN114529739A (zh) | 一种图像匹配方法 |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 20956564 Country of ref document: EP Kind code of ref document: A1 |