WO2017084261A1 - 一种用于图像配准的图像预处理方法及装置 - Google Patents

一种用于图像配准的图像预处理方法及装置 Download PDF

Info

Publication number
WO2017084261A1
WO2017084261A1 PCT/CN2016/082536 CN2016082536W WO2017084261A1 WO 2017084261 A1 WO2017084261 A1 WO 2017084261A1 CN 2016082536 W CN2016082536 W CN 2016082536W WO 2017084261 A1 WO2017084261 A1 WO 2017084261A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
registered
filtered
original image
registration
Prior art date
Application number
PCT/CN2016/082536
Other languages
English (en)
French (fr)
Inventor
张修宝
Original Assignee
乐视控股（北京）有限公司
乐视致新电子科技（天津）有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 乐视控股（北京）有限公司, 乐视致新电子科技（天津）有限公司 filed Critical 乐视控股（北京）有限公司
Priority to US15/243,211 priority Critical patent/US20170140538A1/en
Publication of WO2017084261A1 publication Critical patent/WO2017084261A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20048—Transform domain processing
    • G06T2207/20056—Discrete and fast Fourier transform, [DFT, FFT]

Definitions

  • Embodiments of the present invention relate to the field of image processing, and in particular, to an image preprocessing method and apparatus for image registration.
  • Image registration is the process of matching and superimposing two or more images acquired at different times, different sensors (imaging devices) or under different conditions (weather, illumination, camera position and angle, etc.). It is widely used in computer vision, image processing, remote sensing data analysis, image fusion, image super-resolution reconstruction and medical image processing.
  • Image registration can be divided into two categories according to the method used: region-based image registration and feature-based image registration.
  • the theoretical basis of the mutual power spectrum (phase correlation) registration method is the Fourier transform.
  • FFT fast Fourier algorithm
  • the registration method has the advantages of simple algorithm and fast speed. It has a wide range of applications in image registration, pattern recognition, and feature matching.
  • the image registration method based on the cross-power spectrum mainly utilizes the variation of high-frequency components in the image, and the high-frequency components are susceptible to noise, which leads to a decrease in the accuracy of image registration.
  • Embodiments of the present invention provide an image preprocessing method and apparatus for image registration, which are used to solve defects caused by noise during image registration, and improve image registration accuracy.
  • An embodiment of the present invention provides an image preprocessing method for image registration, including:
  • the window size of the Gaussian filter and the smoothness parameter of the Gaussian filter are selected to construct a Gaussian filter function.
  • the first-order partial derivatives of the different directions are respectively convoluted with the original image in the corresponding direction and the image to be registered in the corresponding direction, to obtain the original image filtered by the corresponding direction and the filtered image to be registered;
  • An embodiment of the present invention provides an image preprocessing apparatus for image registration, including:
  • a setting module is configured to select a window size of the Gaussian filter and a smoothness parameter of the Gaussian filter to construct a Gaussian filter function.
  • a calculation module configured to separately calculate first-order partial derivatives of different directions of the Gaussian filter function
  • a filtering module configured to convolute the first-order partial derivatives of the different directions with the original image in the corresponding direction and the image to be registered in the corresponding direction, to obtain the original image filtered by the corresponding direction and the filtered to-be-matched Quasi-image
  • a registration module configured to perform image registration on the filtered original image and the filtered image to be registered.
  • the image preprocessing method and apparatus for image registration provided by the embodiment of the present invention performs convolution preprocessing on the image by using a Gaussian kernel, that is, filtering the image, which can effectively eliminate the part existing in the image with less change in content.
  • a Gaussian kernel that is, filtering the image, which can effectively eliminate the part existing in the image with less change in content.
  • the influence of noise, highlighting the richness of the details in the image, and the contribution of the larger content to the image registration improves the accuracy of image registration.
  • Embodiment 1 is a technical flowchart of Embodiment 1 of the present invention.
  • Embodiment 2 is a technical flowchart of Embodiment 2 of the present invention.
  • Embodiment 3 is a technical flowchart of Embodiment 3 of the present invention.
  • FIG. 4 is a schematic structural diagram of a device according to Embodiment 4 of the present invention.
  • FIG. 1 is a technical flowchart of Embodiment 1 of the present invention.
  • an image preprocessing method for image registration of the present invention is mainly implemented by the following steps:
  • Step 110 Select a Gaussian filter window size and a Gaussian filter smoothness parameter to construct a Gaussian filter function
  • Image filtering generally includes spatial domain filtering and frequency domain filtering. If the output pixel is a linear combination of input pixel neighborhood pixels, it is called linear filtering (such as the most common mean filtering and Gaussian filtering). Wave), otherwise nonlinear filtering (median filtering, edge-preserving filtering, etc.).
  • linear filtering such as the most common mean filtering and Gaussian filtering. Wave
  • nonlinear filtering median filtering, edge-preserving filtering, etc.
  • the linear smoothing filter works well for removing Gaussian noise and, in most cases, works well for other types of noise.
  • the original image and the image to be registered are not directly filtered by the Gaussian filter, but are filtered by convolution of the Gaussian derivative kernel, thereby better extracting the image while denoising Change information in all directions.
  • the Gaussian derivative kernel convolution utilizes the partial differential of the Gaussian function. Therefore, it is necessary to preset the parameters of the Gaussian function, that is, select the window size and smoothing parameters of the Gaussian filter.
  • the filtering of the Gaussian filter is a process of weighted averaging of the entire image.
  • the value of each pixel is obtained by weighted averaging of itself and other pixel values in the neighborhood of the window.
  • the selection of the window size is critical. If the window size is too small, the denoising of the pixel is not enough, and the pixel is susceptible to noise; the window size is too large, which increases the amount of calculation.
  • the embodiment of the present invention generally selects a window size of 7*7, which is an empirical value. Generally, setting the window of the Gaussian filter to 7*7 can achieve the best filtering effect, but the embodiment of the present invention does not measure the size of the window. Make restrictions.
  • Step 120 respectively calculate first-order partial derivatives of different directions of the Gaussian filter function
  • the different directions are determined according to the dimensions of the processed picture.
  • the mathematical model corresponding to the two-dimensional Gaussian filter should be selected to calculate the partial derivative.
  • the mathematical model of the two-dimensional Gaussian filter is as follows:
  • the width of the Gaussian filter (determining the degree of smoothing) is represented by the smoothing parameter ⁇ , and the relationship between the smoothing parameter ⁇ and the degree of smoothing is very simple, and the larger the ⁇ , the wider the frequency band of the Gaussian filter. The smoother the better.
  • Gx is the first-order partial derivative of the two-dimensional Gaussian function G(x, y) in the horizontal direction (x direction)
  • Gy is the one of the two-dimensional Gaussian function G(x, y) in the vertical direction (y direction) Order derivative
  • Step 130 Convolve the first-order partial derivatives of the different directions with the original image in the corresponding direction and the image to be registered in the corresponding direction, and obtain the original image filtered by the corresponding direction and the filtered image to be registered. ;
  • the filtered result of the Gaussian derivative kernel convolution is presented as a gradient image, and the gradient image of the two-dimensional Gaussian derivative kernel convolution is as follows:
  • I(x, y) is the function corresponding to the image to be filtered
  • I_Gx(x, y) is the horizontally filtered gradient image
  • I_Gy(x, y) is the filtered gradient in the vertical direction. image.
  • the gradient images corresponding to the horizontal direction and the vertical direction corresponding to the original image after filtering are respectively:
  • the filtered gradient images corresponding to the horizontal direction and the vertical direction corresponding to the image to be registered are respectively:
  • Step 140 Perform image registration on the filtered original image and the filtered image to be registered.
  • the image preprocessing method of the embodiment of the present invention is applied to an image registration method based on a mutual power spectrum.
  • the image registration method based on cross-power spectrum is a kind of image registration method based on region, which is detected by translation, rotation and scaling between two images to achieve fast registration of images.
  • the image registration method based on mutual power spectrum mainly utilizes the change of high frequency components after image transformation. It usually directly calculates the registration of two images, which includes both low frequency components and high frequency components, but due to high The frequency components are susceptible to noise and the accuracy of the registration is reduced.
  • the displacement theory guarantees the equivalence between the phase of the cross power spectrum and the phase difference between the two images. Therefore, by performing the inverse Fourier transform on the cross power spectrum, the pulse function ⁇ (x-x0, y-y0) can be obtained. Since the pulse function has a sharp sharp peak at the offset position (x0, y0), the values of the other positions are close to zero, whereby the offset between the two images can be obtained.
  • r(x, y) be a two-dimensional image whose Fourier transform is R(u,v), and the displacement of the image p(x,y) relative to the image r(x,y) occurs (x 0 ,y 0 ) :
  • R -1 ⁇ represents the inverse Fourier transform
  • the pulse function ⁇ (x-x0, y-y0) has a sharp sharp peak at the offset position (x0, y0), and the values of other positions are close to zero, whereby the offset between the two images can be obtained.
  • the original image and the image to be registered after filtering are respectively subjected to horizontal registration and vertical registration according to the above principle, that is, fh_Gx(x, y) and gh_Gx(x, y) are registered, Then gh_Gx(x,y) and gv_Gy(x,y) are registered.
  • fh_Gx(x, y) and gh_Gx(x, y) are registered.
  • gh_Gx(x,y) and gv_Gy(x,y) are registered.
  • the order of the above registration is only an example, and the registration in the horizontal direction and the registration in the vertical direction are actually not in the order. Since image registration is very mature in the prior art, it will not be described here.
  • the image is filtered and preprocessed by Gaussian kernel convolution, which effectively eliminates the influence of noise existing in the portion where the content changes little in the image, and highlights the image with rich details and the content change is large.
  • Gaussian kernel convolution effectively eliminates the influence of noise existing in the portion where the content changes little in the image, and highlights the image with rich details and the content change is large.
  • the contribution of the high-frequency component enhances the accuracy of image registration.
  • Embodiment 2 is a technical flow chart of Embodiment 2 of the present invention. The following part will be combined with FIG. 2 to explain a shift amount between images in an image preprocessing method for image registration according to an embodiment of the present invention in a more detailed embodiment. The implementation steps of the registration.
  • Step 210 Select a Gaussian filter window size and a Gaussian filter smoothness parameter to construct a Gaussian filter function
  • Step 220 respectively calculate first-order partial derivatives of different directions of the Gaussian filter function
  • Step 230 Perform boundary expansion on the original image and the image to be registered to obtain the original image and the image to be registered that are adapted to the window size of the Gaussian filter.
  • the Gaussian function-based filtering method uses window neighboring for each pixel in the filtering process
  • the pixel value of the pixel in the domain is weighted and averaged, so the boundary of the image is expanded to facilitate processing of the pixel of the edge portion of the image.
  • the boundary of the original image needs to be extended outward by at least two pixels, and the pixel on the edge of the original image can be denoised by weighting and averaging the pixels in the neighborhood.
  • the image boundary expansion method used in the embodiment of the present invention may be filled with values of padding zero, period padding, mirror padding, or copy outer boundary.
  • the original image is subjected to boundary expansion by using a mirror filling method, and the size of the expansion is determined according to the selected filtering window size.
  • the embodiment of the present invention can directly implement boundary expansion of an image by using OpenCV.
  • OpenCV provides several different boundary extension strategies:
  • src is the input array
  • dst is the array after the extended boundary of the output
  • top The number of rows that expand upward on the src boundary
  • bottom is the number of rows that expand downwards on the lower boundary of src
  • left is the number of columns that extend to the left on the left edge of src
  • right is the column that extends to the right on the right edge of src Number
  • borderType is one of the boundary expansion strategies
  • value is the constant filled in the boundary when BORDER_CONSTANT is used as the boundary expansion strategy value.
  • the image may be extended first, and then the first-order partial derivative of the Gaussian filter function in different directions may be calculated.
  • Step 240 Convolve the first-order partial derivatives of the different directions with the original image in the corresponding direction and the image to be registered in the corresponding direction, and obtain the original image filtered by the corresponding direction and the filtered image to be registered. ;
  • step 240 further includes step 241 and step 242.
  • Step 241 Convolution operation of the partial derivative in the horizontal direction with the original image and the image to be registered to obtain a horizontally filtered original image and a horizontally filtered image to be registered;
  • Step 242 Convolve the partial derivative of the vertical direction with the original image and the image to be registered to obtain a vertically filtered original image and a vertically filtered image to be registered.
  • Step 250 Perform boundary clipping on the filtered original image and the filtered to-be-registered image, where the boundary-cut region is an area where the boundary is expanded.
  • the original image and the image to be registered are both extended to obtain an image size matching the Gaussian filtering window, and the main purpose is to conveniently process the pixel points of the original edge portion of the image, but the actual registration is performed. This is not required for this extended part. Therefore, preferably, in the embodiment of the present invention, after the filtering process is completed, the area that does not belong to the original image needs to be cropped to ensure the registration effect.
  • Step 260 Perform registration of the inter-image shift amount between the filtered original image and the filtered image to be registered.
  • the original image and the image to be processed are boundary-expanded for the purpose of obtaining a better filtering effect; and the original image and the image to be processed are filtered to achieve accurate image-to-image flatness. Registration of the shift amount.
  • Embodiment 3 is a technical flow chart of Embodiment 3 of the present invention. The following part will be described in conjunction with FIG. 3 to explain a rotation angle between images in an image preprocessing method for image registration according to a more detailed embodiment of the present invention. The implementation steps of the registration.
  • Step 310 Perform polar coordinate transformation on the original image and the image to be registered to obtain the original image and the image to be registered in polar coordinates;
  • the rotation angle between the original image and the image to be registered is first obtained.
  • the original image and the image to be registered are first subjected to polar coordinate transformation.
  • the amount of translation of the original image and the image to be registered in polar coordinates is obtained, and the translation amount in polar coordinates is converted into a rotation angle value in a plane rectangular coordinate system.
  • Step 320 Select a window size of the Gaussian filter and a smoothness parameter of the Gaussian filter to construct a Gaussian filter function
  • Step 330 Calculate first-order partial derivatives of different directions of the Gaussian filter function respectively;
  • Step 340 Perform boundary expansion on the original image and the image to be registered to obtain the original image and the image to be registered that are adapted to the window size of the Gaussian filter.
  • step 330 there is no sequence between step 330 and step 340.
  • the image may be extended first, and then the first-order partial derivative of the Gaussian filter function in different directions may be calculated.
  • Step 350 Convolve the first-order partial derivatives of the different directions with the original image in the corresponding direction and the image to be registered in the corresponding direction, and obtain the original image filtered by the corresponding direction and the filtered image to be registered. ;
  • step 340 further includes step 341 and step 342.
  • Step 351 is to divide the partial derivative of the horizontal direction with the original image and the to-be-matched
  • the quasi-image is subjected to a convolution operation to obtain a horizontally filtered original image and a horizontally filtered image to be registered;
  • Step 352 is to convolve the partial derivative of the vertical direction with the original image and the image to be registered to obtain a vertically filtered original image and a vertically filtered image to be registered.
  • Step 360 Perform boundary clipping on the filtered original image and the filtered to-be-registered image, where the boundary-cut region is an area where the boundary is expanded.
  • Step 370 Acquire a translation amount of the filtered original image and the filtered image to be registered in polar coordinates, and then convert the polar coordinates into rectangular coordinates to obtain the original image and the to-be-registered image. The angle of rotation between.
  • the embodiment of the invention adopts image registration based on Fourier-Merlin transform, and the principle is as follows:
  • the image f2(x, y) is the translation of the image f1(x, y) (x 0 , y 0 ), and the result after rotating the angle ⁇ 0 , ie
  • f 2 (x, y) f 1 (xcos ⁇ 0 + ysin ⁇ 0 -x 0, -xsin ⁇ 0 + ycos ⁇ 0 -y 0)
  • M 2 ( ⁇ , ⁇ ) M 1 ( ⁇ cos ⁇ 0 + ⁇ sin ⁇ 0 , - ⁇ sin ⁇ 0 + ⁇ cos ⁇ 0 )
  • the amount of translation in the direction of the angular axis in the polar coordinate system is obtained, and then the rotation angle is obtained by converting the polar coordinate to the rectangular coordinate.
  • the translation of the original image and the image to be registered in polar coordinates is obtained.
  • an image pre-processing device for image registration mainly includes the following modules: a setting module 410, a calculation module 420, a filtering module 430, and a matching module. Quasi-module 440.
  • the setting module 410 is configured to select a window size of the Gaussian filter and a smoothness parameter of the Gaussian filter to construct a Gaussian filter function.
  • the calculation module 420 is connected to the setting module 410, and configured to respectively calculate first-order partial derivatives of different directions of the Gaussian filter function according to parameters set by the setting module 410;
  • the filtering module 430 is connected to the calculating module 420, and is configured to perform convolution operations on the first-order partial derivatives of the different directions and the original image in the corresponding direction and the image to be registered in the corresponding direction to obtain corresponding direction filtering. After the original image and the filtered image to be registered;
  • the registration module 440 is configured to perform image registration on the filtered original image and the filtered image to be registered.
  • the device further includes an image expansion module 450, and the image expansion module 450 is configured to perform convolution operation on the first-order partial derivatives of the different directions and the original image in the corresponding direction and the image to be registered in the corresponding direction.
  • the original image and the image to be registered are subjected to boundary expansion to obtain the original image and the image to be registered which are adapted to the window size of the Gaussian filter.
  • the filtering module 430 is configured to convolute the partial derivative in the horizontal direction with the original image and the image to be registered to obtain a horizontally filtered original image and a horizontally filtered candidate to be matched. a quasi-image; convolving the partial derivative in the vertical direction with the original image and the image to be registered to obtain a vertically filtered original image and vertical filtering The image to be registered after.
  • the apparatus further includes an image cropping module 460 for
  • the apparatus further includes a coordinate transformation module 470, configured to pre-position the original image and the image to be registered before performing the convolution operation when performing registration of an image rotation angle
  • the polar coordinate transformation is performed to obtain the original image and the image to be registered in polar coordinates.
  • the device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separate, and the components displayed as units may or may not be physical units, ie may be located A place, or it can be distributed to multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the embodiment. Those of ordinary skill in the art can understand and implement without deliberate labor.

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Image Processing (AREA)

Abstract

一种用于图像配准的图像预处理方法及装置。选取高斯滤波的窗口尺寸和高斯滤波的平滑度参数用以构造高斯滤波函数(110)；分别计算所述高斯滤波函数不同方向的一阶偏导数(120)；将所述不同方向的一阶偏导数分别与相应方向的原图像和相应方向的待配准的图像进行卷积运算，得到相应方向滤波后的原图像和滤波后的待配准图像(130)；将所述滤波后的原图像和所述滤波后的待配准图像进行图像配准(140)。实现了图像配准之间的滤波去噪，提高了图像配准的精度。

Description

一种用于图像配准的图像预处理方法及装置
交叉引用
本申请引用于2015年11月16日递交的名称为“一种用于图像配准的图像预处理方法及装置”的第2015107864254号中国专利申请，其通过引用被全部并入本申请。
技术领域
本发明实施例涉及图像处理领域，尤其涉及一种用于图像配准的图像预处理方法及装置。
背景技术
图像配准(Image registration)就是将不同时间、不同传感器(成像设备)或不同条件下(天候、照度、摄像位置和角度等)获取的两幅或多幅图像进行匹配、叠加的过程，它已经被广泛地应用于计算机视觉、图像处理、遥感数据分析、图像融合、图像的超分辨率重建和医学图像处理等领域。
按照所使用的方法，图像配准可以分为两类：基于区域的图像配准和基于特征的图像配准。基于互功率谱(相位相关)配准方法的理论基础是傅里叶变换，在傅里叶变换领域有了快速傅里叶算法FFT的前提下，该配准方法具有算法简单，速度快等优势，在图像配准、模式识别、特征匹配等有着广泛应用。
但是，基于互功率谱的图像配准方法主要利用了图像中高频成分的变化，而高频成分易受噪声影响，因而导致图像配准的精度会降低。
因此，一种用于图像配准的图像预处理方法亟待提出。
发明内容
本发明实施例提供一种用于图像配准的图像预处理方法及装置，用以解决图像配准过程中因受噪声影响的缺陷，提高了图像配准的精确度。
本发明实施例提供一种用于图像配准的图像预处理方法，包括：
选取高斯滤波的窗口尺寸和高斯滤波的平滑度参数用以构造高斯滤波函数。
分别计算所述高斯滤波函数不同方向的一阶偏导数；
将所述不同方向的一阶偏导数分别与相应方向的原图像和相应方向的待配准的图像进行卷积运算，得到相应方向滤波后的原图像和滤波后的待配准图像；
将所述滤波后的原图像和所述滤波后的待配准图像进行图像配准。
本发明实施例提供一种用于图像配准的图像预处理装置，包括：
设置模块，用于选取高斯滤波的窗口尺寸和高斯滤波的平滑度参数用以构造高斯滤波函数。
计算模块，用于分别计算所述高斯滤波函数不同方向的一阶偏导数；
滤波模块，用于将所述不同方向的一阶偏导数分别与相应方向的原图像和相应方向的待配准的图像进行卷积运算，得到相应方向滤波后的原图像和滤波后的待配准图像；
配准模块，用于将所述滤波后的原图像和所述滤波后的待配准图像进行图像配准。
本发明实施例提供的用于图像配准的图像预处理方法和装置通过高斯核对图像进行卷积预处理，即对图像进行滤波处理，可以有效消除存在于图像中内容变化较小的部分中的噪声的影响，突出图像中细节丰富，内容变化较大的部分对图像配准的贡献，提高了图像配准的精度。
附图说明
此处所说明的附图用来提供对本发明的进一步理解，构成本申请的一部分，本发明的示意性实施例及其说明用于解释本发明，并不构成对本发明的不当限定。在附图中：
图1为本发明实施例一的技术流程图；
图2为本发明实施例二的技术流程图；
图3为本发明实施例三的技术流程图；
图4为本发明实施例四的装置结构示意图。
具体实施方式
为使本发明实施例的目的、技术方案和优点更加清楚，下面将结合本发明实施例中的附图，对本发明实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例是本发明一部分实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有作出创造性劳动前提下所获得的所有其他实施例，都属于本发明保护的范围。
需要说明的是，本发明的各个实施例之间不是独立存在的，而是一种技术方案在不同应用场景的详细阐述。
实施例一
图1是本发明实施例一的技术流程图，结合图1，本发明一种用于图像配准的图像预处理方法主要由如下的步骤实现：
步骤110：选取高斯滤波的窗口尺寸和高斯滤波的平滑度参数用以构造高斯滤波函数；
图像滤波总体上讲包括空域滤波和频域滤波。如果输出像素是输入像素邻域像素的线性组合则称为线性滤波(例如最常见的均值滤波和高斯滤 波)，否则为非线性滤波(中值滤波、边缘保持滤波等)。线性平滑滤波器去除高斯噪声的效果很好，且在大多数情况下，对其它类型的噪声也有很好的效果。
本发明实施例中，对原图像和待配准图像并不直接通过高斯滤波器进行滤波，而是通过高斯导数核的卷积进行滤波，由此可以在去噪的同时，更好地提取图像中各个方向的变化信息。高斯导数核卷积利用了高斯函数的偏微分，因此，还需预先设置高斯函数的参数，即选取高斯滤波器的窗口尺寸和平滑参数。
高斯滤波器的滤波是对整幅图像进行加权平均的过程，每一个像素点的值，都由其本身和窗口邻域内的其他像素值经过加权平均后得到。窗口尺寸的选择很关键，窗口尺寸太小，则对于像素点的去噪程度不够，像素点容易受到噪声的影响；窗口尺寸太大，会增加计算量。本发明实施例一般选用7*7的窗口大小，这是一个经验值，通常将高斯滤波器的窗口设置为7*7能够达到最好的滤波效果，但是本发明实施例并不对窗口尺寸的大小做限制。
步骤120：分别计算所述高斯滤波函数不同方向的一阶偏导数；
本发明实施例中，所述不同方向是根据被处理的图片的维度确定的，例如，对于普通的二维平面图片，应当选取二维高斯滤波器对应的数学模型进行偏导数的计算。二维高斯滤波器的数学模型如下：
Figure PCTCN2016082536-appb-000001
其中，σ是平滑参数，高斯滤波器宽度(决定着平滑程度)是由平滑参数σ表征的，而且平滑参数σ和平滑程度的关系非常简单，σ越大，高斯滤波器的频带就越宽，平滑程度就越好。
当本发明实施例的应用场景为二维图像的图像配准时，分别计算二维 高斯函数在水平和垂直两个方向的一阶偏导数：
Figure PCTCN2016082536-appb-000002
Figure PCTCN2016082536-appb-000003
其中，Gx为二维高斯函数G(x,y)在水平方向(x方向)上的一阶偏导数，Gy为二维高斯函数G(x,y)在垂直方向(y方向)上的一阶偏导数，
步骤130：将所述不同方向的一阶偏导数分别与相应方向的原图像和相应方向的待配准的图像进行卷积运算，得到相应方向滤波后的原图像和滤波后的待配准图像；
高斯导数核卷积的滤波结果以梯度图像展现，二维高斯导数核卷积的梯度图像如下所示：
Figure PCTCN2016082536-appb-000004
Figure PCTCN2016082536-appb-000005
其中，
Figure PCTCN2016082536-appb-000006
表示卷积运算，I(x,y)是待滤波的图像对应的函数，I_Gx(x,y)是水平方向上滤波后的梯度图像，I_Gy(x,y)是垂直方向上滤波后的梯度图像。
本发明实施例中，假设原图像为f(x,y)，待配准图像为g(x,y)，则滤波后所述原图像对应的水平方向和垂直方向对应的梯度图像分别为：
Figure PCTCN2016082536-appb-000007
Figure PCTCN2016082536-appb-000008
滤波后所述待配准图像对应的水平方向和垂直方向对应的梯度图像分别为：
Figure PCTCN2016082536-appb-000009
Figure PCTCN2016082536-appb-000010
步骤140：将所述滤波后的原图像和所述滤波后的待配准图像进行图像配准。
优选地，将本发明实施例的图像预处理方法应用于基于互功率谱的图像配准方法中。基于互功率谱(相位相关)的图像配准方法是基于区域的图像配准方法的一种，通过两幅图像间的平移、旋转、缩放进行检测，实现对图像的快速配准。基于互功率谱的图像配准方法主要利用了图像变换后的高频成分的变化，它通常直接对两幅图像进行配准计算，这其中既包含低频成分，也包含高频成分，但由于高频成分易受噪声影响，因而配准的精度会降低。
位移理论保证了互功率谱的相位与两幅图像相位差的等效性，因此通过对互功率谱进行傅里叶逆变换，可以得到脉冲函数δ(x-x0,y-y0)。由于脉冲函数在偏移位置(x0,y0)处有明显的尖锐峰值，其它位置的值接近于零，据此即能获取两图像间的偏移量。
基于互功率谱的图像配准原理如下：
设r(x,y)为二维图像，其傅里叶变换为R(u,v)，图像p(x,y)相对图像r(x,y)发生(x0,y0)的位移：
p(x,y)＝r(x-x0,y-y0)
p(x,y)的傅里叶变换为：
Figure PCTCN2016082536-appb-000011
于是，其归一化的互功率谱可以表示为：
Figure PCTCN2016082536-appb-000012
其中，P*(u,v)表示P(u,v)的共轭。
Figure PCTCN2016082536-appb-000013
其中，R-1{}表示傅里叶逆变换。
脉冲函数δ(x-x0,y-y0)在偏移位置(x0,y0)处有明显的尖锐峰值，其它位置的值接近于零，据此即能获取两图像间的偏移量。
本发明实施例中，分别对于滤波处理后的原图像和待配准图像分别按照上述原理进行水平配准以及垂直配准，即将fh_Gx(x,y)和gh_Gx(x,y)进行配准，再将gh_Gx(x,y)和gv_Gy(x,y)进行配准。当然，上述配准的顺序仅供举例，水平方向的配准和垂直方向的配准实际上并无先后顺序。因图像配准在现有技术中十分成熟，此处不再赘述。
本实施例中，通过高斯核卷积对图像进行滤波预处理，有效消除存在于图像中内容变化较小的部分中噪声的影响，突出图像中细节丰富，内容变化较大的部分对图像配准的贡献，增强高频成份的作用，从而提高了图像配准的精度。
实施例二
图2是本发明实施例二的技术流程图，以下部分将结合图2，以一个更加详细的实施例说明本发明实施例一种用于图像配准的图像预处理方法中，图像间平移量配准的实现步骤。
步骤210：选取高斯滤波的窗口尺寸和高斯滤波的平滑度参数用以构造高斯滤波函数；
步骤220：分别计算所述高斯滤波函数不同方向的一阶偏导数；
步骤230：对原图像和待配准的图像进行边界扩充以得到与高斯滤波的所述窗口尺寸大小相适应的所述原图像和所述待配准的图像。
因基于高斯函数的滤波方法在滤波处理时，对每个像素点采用窗口邻 域内的像素点的像素值加权求平均的算法，因此对图像进行边界进行扩展方便处理图像边缘部分的像素点。例如，对于窗口为3*3的高斯滤波，需要将原图片的边界向外扩展至少两个像素点，能够对原图片边缘上的像素点采用邻域内像素点加权求平均的方式进行去噪。
本发明实施例中使用的图像边界扩展方法可以是填充零、周期填充、镜像填充或复制外边界的值进行填充。
优选地，本发明实施例中采用镜像填充的方法对原图像进行边界扩展，扩展的大小根据选择的滤波窗口尺寸决定
可选地，本发明实施例还可以借助OpenCV直接实现图像的边界扩展。OpenCV提供了几种不同的边界扩展策略：
*BORDER_REPLICATE:aaaaaa|abcdefgh|hhhhhhh
*BORDER_REFLECT:fedcba|abcdefgh|hgfedcb
*BORDER_REFLECT_101:gfedcb|abcdefgh|gfedcba
*BORDER_WRAP:cdefgh|abcdefgh|abcdefg
*BORDER_CONSTANT:iiiiii|abcdefgh|iiiiiii with some specified’i’
其中“|”表示的是图像的边界，连个“|”中间是图像的内容，最后一个边界扩展策略还要额外给定一个i值，用于对额外的边界进行赋值。
使用OpenCV提供的函数copyMakeBorder()来扩展边界，其原型如下：
void copyMakeBorder(InputArray src,OutputArray dst,int top,int bottom,int left,int right,int borderType,const Scalar&value＝Scalar())其中，src为输入的数组；dst为输出的扩展边界后的数组；top为在src上边界向上扩展的行数；bottom为在src下边界向下扩展的行数；left为在src的左边界向左扩展的列数；right为在src的右边界向右扩展的列数；borderType为边界扩展策略中的一个；value为当使用BORDER_CONSTANT作为所述边界扩展策略的时候，边界处填写的常数 值。
需要说明的是，步骤220和步骤230之间没有先后顺序，本发明实施例中也可以是先对图像进行扩展，再计算高斯滤波函数不同方向的一阶偏导数。
步骤240：将所述不同方向的一阶偏导数分别与相应方向的原图像和相应方向的待配准的图像进行卷积运算，得到相应方向滤波后的原图像和滤波后的待配准图像；
具体地，步骤240进一步包括步骤241和步骤242。
步骤241：将水平方向的所述偏导数分别与所述原图像和所述待配准的图像进行卷积运算得到水平滤波后的原图像和水平滤波后的待配准图像；
步骤242：将垂直方向的所述偏导数分别与所述原图像和所述待配准的图像进行卷积运算得到垂直滤波后的原图像和垂直滤波后的待配准图像。
步骤250：对所述滤波后的原图像和所述滤波后的待配准图像进行边界裁剪，其中，所述边界裁剪的区域为所述边界扩充的区域。
本发明实施例中，将原图像和待配准的图像均进行边界扩展以获得与高斯滤波窗口相匹配的图像大小，主目的是滤波时方便处理图片原来边缘部分的像素点，但实际配准中并不需要这被扩展的部分。因此，优选地，本发明实施例在做完滤波处理后需将此部分本不属于原来图像的区域进行裁剪以保证配准效果。
步骤260：将所述滤波后的原图像和所述滤波后的待配准图像进行图像间平移量的配准。
本实施中对原图像和待处理图像进行边界扩展，目的在于得到更佳的滤波效果；通过对原图像和待处理图像进行滤波，实现了精确的图像间平 移量的配准。
实施例三
图3是本发明实施例三的技术流程图，以下部分将结合图3，以一个更加详细的实施例说明本发明实施例一种用于图像配准的图像预处理方法中，图像间旋转角配准的实现步骤。
步骤310：将所述原图像和所述待配准的图像进行极坐标变换，得到极坐标下的所述原图像和所述待配准的图像；
在进行图像间旋转角的配准时，首先要获取原图像与待配准图像之间的旋转角大小，本发明实施例中采用的方法是先将原图像与待配准图像进行极坐标变换，获取原图像与待配准图像在极坐标下的平移量，再将极坐标下的所述平移量转化为平面直角坐标系下的旋转角度值。
步骤320：选取高斯滤波的窗口尺寸和高斯滤波的平滑度参数用以构造高斯滤波函数；
步骤330：分别计算所述高斯滤波函数不同方向的一阶偏导数；
步骤340：对原图像和待配准的图像进行边界扩充以得到与高斯滤波的所述窗口尺寸大小相适应的所述原图像和所述待配准的图像。
需要说明的是，步骤330和步骤340之间没有先后顺序，本发明实施例中也可以是先对图像进行扩展，再计算高斯滤波函数不同方向的一阶偏导数。
步骤350：将所述不同方向的一阶偏导数分别与相应方向的原图像和相应方向的待配准的图像进行卷积运算，得到相应方向滤波后的原图像和滤波后的待配准图像；
具体地，步骤340进一步包括步骤341和步骤342。
步骤351为将水平方向的所述偏导数分别与所述原图像和所述待配 准的图像进行卷积运算得到水平滤波后的原图像和水平滤波后的待配准图像；
步骤352为将垂直方向的所述偏导数分别与所述原图像和所述待配准的图像进行卷积运算得到垂直滤波后的原图像和垂直滤波后的待配准图像。
步骤360：对所述滤波后的原图像和所述滤波后的待配准图像进行边界裁剪，其中，所述边界裁剪的区域为所述边界扩充的区域。
步骤370：获取所述滤波后的原图像和所述滤波后的待配准图像在极坐标下的平移量，再将极坐标转化为直角坐标，获取所述原图像与所述待配准图像之间的所述旋转角。
优选的，在进行图像间旋转角的配准时，本发明实施例采用基于傅里叶-梅林变换的图像配准，其原理如下所述：
假设图像f2(x,y)为图像f1(x,y)平移(x0,y0)，旋转θ0角后的结果，即
f2(x,y)＝f1(xcosθ0+ysinθ0-x0,-xsinθ0+ycosθ0-y0)
两者傅里叶变换后的关系为
Figure PCTCN2016082536-appb-000014
其幅值M1和M2的关系为
M2(ε,η)＝M1(εcosθ0+ηsinθ0,-εsinθ0+ηcosθ0)
在极坐标系下，两者的关系为
M2(ρ,θ)＝M1(ρ,θ-θ0)
通过计算两者的互功率谱，获得在极坐标系中角度轴方向的平移量，再经转极坐标转换至直角坐标即可获取其旋转角。
本实施例中，通过将平面坐标中的图像转化为极坐标中的图像，再对极坐标中的图像进行滤波去噪，获取极坐标下原图像和待配准图像的平移 量，再通过将极坐标转换到直角坐标，即获得去噪后原图像和待配准图像间的旋转角度值，从而提高了图像旋转角配准的精度。
实施例四
图4是本发明实施例四的装置结构示意图，结合图4，发明实施例一种用于图像配准的图像预处理装置主要包括如下模块：设置模块410、计算模块420、滤波模块430、配准模块440。
所述设置模块410，用于选取高斯滤波的窗口尺寸和高斯滤波的平滑度参数用以构造高斯滤波函数。
所述计算模块420与所述设置模块410相连接，用于根据所述设置模块410设置的参数分别计算所述高斯滤波函数不同方向的一阶偏导数；
所述滤波模块430与所述计算模块420相连，用于将所述不同方向的一阶偏导数分别与相应方向的原图像和相应方向的待配准的图像进行卷积运算，得到相应方向滤波后的原图像和滤波后的待配准图像；
所述配准模块440，用于将所述滤波后的原图像和所述滤波后的待配准图像进行图像配准。
所述装置进一步包括图像扩展模块450，所述图像扩展模块450用于，在将所述不同方向的一阶偏导数分别与相应方向的原图像和相应方向的待配准的图像进行卷积运算之前，对原图像和待配准的图像进行边界扩充以得到与高斯滤波的所述窗口尺寸大小相适应的所述原图像和所述待配准的图像。
进一步地，所述滤波模块430用于将水平方向的所述偏导数分别与所述原图像和所述待配准的图像进行卷积运算得到水平滤波后的原图像和水平滤波后的待配准图像；将垂直方向的所述偏导数分别与所述原图像和所述待配准的图像进行卷积运算得到垂直滤波后的原图像和垂直滤波 后的待配准图像。
所述装置进一步包括图像裁剪模块460，所述图像裁剪模块460用于为
在将所述滤波后的原图像和所述滤波后的待配准图像进行图像配准之前，对所述滤波后的原图像和所述滤波后的待配准图像进行边界裁剪，其中，所述边界裁剪的区域为所述边界扩充的区域。
所述装置进一步包括坐标变换模块470，所述坐标变换模块470用于当进行图像旋转角的配准时，在进行所述卷积运算之前，预先将所述原图像和所述待配准的图像进行极坐标变换，得到极坐标下的所述原图像和所述待配准的图像。
以上所描述的装置实施例仅仅是示意性的，其中所述作为分离部件说明的单元可以是或者也可以不是物理上分开的，作为单元显示的部件可以是或者也可以不是物理单元，即可以位于一个地方，或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部模块来实现本实施例方案的目的。本领域普通技术人员在不付出创造性的劳动的情况下，即可以理解并实施。
上述说明示出并描述了本申请的实施例，但如前所述，应当理解本申请并非局限于本文所披露的形式，不应看作是对其他实施例的排除，而可用于各种其他组合、修改和环境，并能够在本文所述发明构想范围内，通过上述教导或相关领域的技术或知识进行改动。而本领域人员所进行的改动和变化不脱离本申请的精神和范围，则都应在本申请所附权利要求的保护范围内。

Claims (10)

  1. 一种用于图像配准的图像预处理方法，其特征在于，包括如下的步骤：
    选取高斯滤波的窗口尺寸和高斯滤波的平滑度参数用以构造高斯滤波函数；
    分别计算所述高斯滤波函数不同方向的一阶偏导数；
    将所述不同方向的一阶偏导数分别与相应方向的原图像和相应方向的待配准的图像进行卷积运算，得到相应方向滤波后的原图像和滤波后的待配准图像；
    将所述滤波后的原图像和所述滤波后的待配准图像进行图像配准。
  2. 根据权利要求1所述的方法，其特征在于，将所述不同方向的一阶偏导数分别与相应方向的原图像和相应方向的待配准的图像进行卷积运算之前，进一步包括：
    对原图像和待配准的图像进行边界扩充以得到与高斯滤波的所述窗口尺寸大小相适应的所述原图像和所述待配准的图像。
  3. 根据权利要求1所述的方法，将所述不同方向的一阶偏导数分别与相应方向的原图像和相应方向的待配准的图像进行卷积运算，进一步包括：
    将水平方向的所述偏导数分别与所述原图像和所述待配准的图像进行卷积运算得到水平滤波后的原图像和水平滤波后的待配准图像；
    将垂直方向的所述偏导数分别与所述原图像和所述待配准的图像进行卷积运算得到垂直滤波后的原图像和垂直滤波后的待配准图像。
  4. 根据权利要求1或2所述的方法，其特征在于，所述方法进一步包括：
    将所述滤波后的原图像和所述滤波后的待配准图像进行图像配准之前，对所述滤波后的原图像和所述滤波后的待配准图像进行边界裁剪，其中，所述边界裁剪的区域为所述边界扩充的区域。
  5. 根据权利要求1所述的方法，其特征在于，所述方法进一步包括：
    当进行图像旋转角的配准时，在进行所述卷积运算之前，预先将所述原图像和所述待配准的图像进行极坐标变换，得到极坐标下的所述原图像和所述待配准的图像。
  6. 一种用于图像配准的图像预处理装置，其特征在于，包括如下的模块：
    设置模块，用于选取高斯滤波的窗口尺寸和高斯滤波的平滑度参数用以构造高斯滤波函数；
    计算模块，用于分别计算所述高斯滤波函数不同方向的一阶偏导数；
    滤波模块，用于将所述不同方向的一阶偏导数分别与相应方向的原图像和相应方向的待配准的图像进行卷积运算，得到相应方向滤波后的原图像和滤波后的待配准图像；
    配准模块，用于将所述滤波后的原图像和所述滤波后的待配准图像进行图像配准。
  7. 根据权利要求6所述的装置，其特征在于，所述装置进一步包括图像扩展模块，所述图像扩展模块用于：
    在将所述不同方向的一阶偏导数分别与相应方向的原图像和相应方向的待配准的图像进行卷积运算之前，对原图像和待配准的图像进行边界扩充以得到与高斯滤波的所述窗口尺寸大小相适应的所述原图像和所述待配准的图像。
  8. 根据权利要求6所述的装置，所述滤波模块进一步用于：
    将水平方向的所述偏导数分别与所述原图像和所述待配准的图像进行 卷积运算得到水平滤波后的原图像和水平滤波后的待配准图像；
    将垂直方向的所述偏导数分别与所述原图像和所述待配准的图像进行卷积运算得到垂直滤波后的原图像和垂直滤波后的待配准图像。
  9. 根据权利要求6或7所述的装置，其特征在于，所述装置进一步包括图像裁剪模块，所述图像裁剪模块用于：
    在将所述滤波后的原图像和所述滤波后的待配准图像进行图像配准之前，对所述滤波后的原图像和所述滤波后的待配准图像进行边界裁剪，其中，所述边界裁剪的区域为所述边界扩充的区域。
  10. 根据权利要求6所述的装置，其特征在于，所述装置进一步包括坐标变换模块，所述坐标变换模块用于：
    当进行图像旋转角的配准时，在进行所述卷积运算之前，预先将所述原图像和所述待配准的图像进行极坐标变换，得到极坐标下的所述原图像和所述待配准的图像。
PCT/CN2016/082536 2015-11-16 2016-05-18 一种用于图像配准的图像预处理方法及装置 WO2017084261A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US15/243,211 US20170140538A1 (en) 2015-11-16 2016-08-22 Image preprocessing method and electronic device for image registration

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201510786425.4 2015-11-16
CN201510786425.4A CN105761236A (zh) 2015-11-16 2015-11-16 一种用于图像配准的图像预处理方法及装置

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US15/243,211 Continuation US20170140538A1 (en) 2015-11-16 2016-08-22 Image preprocessing method and electronic device for image registration

Publications (1)

Publication Number Publication Date
WO2017084261A1 true WO2017084261A1 (zh) 2017-05-26

Family

ID=56342052

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2016/082536 WO2017084261A1 (zh) 2015-11-16 2016-05-18 一种用于图像配准的图像预处理方法及装置

Country Status (2)

Country Link
CN (1) CN105761236A (zh)
WO (1) WO2017084261A1 (zh)

Families Citing this family (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106897694A (zh) * 2017-02-24 2017-06-27 西安天和防务技术股份有限公司 用于国土资源监控的违建场景识别方法
CN107749986B (zh) * 2017-09-18 2019-12-06 深圳市天英联合教育股份有限公司 教学视频生成方法、装置、存储介质和计算机设备
CN111160148A (zh) * 2019-12-16 2020-05-15 上海聪育智能科技有限公司 一种基于像素差异的教学考勤识别系统及方法
CN111127529B (zh) * 2019-12-18 2024-02-02 浙江大华技术股份有限公司 图像配准方法及装置、存储介质、电子装置
CN111815644A (zh) * 2020-05-21 2020-10-23 艾瑞迈迪医疗科技（北京）有限公司 一种医学影像中患者皮肤面部数据的提取方法和装置
CN111739073A (zh) * 2020-06-24 2020-10-02 刘秀萍 高效快速的手持装置的影像配准优化方法
CN114913096B (zh) * 2022-06-10 2024-04-23 中国科学院长春光学精密机械与物理研究所 一种特征初始化的非均匀校正方法及其系统

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101714254A (zh) * 2009-11-16 2010-05-26 哈尔滨工业大学 联合多尺度sift和区域不变矩特征的配准控制点提取方法
CN102208109A (zh) * 2011-06-23 2011-10-05 南京林业大学 X射线图像和激光图像的异源图像配准方法
CN102646264A (zh) * 2011-02-18 2012-08-22 北京卫金帆医学技术发展有限公司 补偿机械运动误差的图像配准方法
US20130301950A1 (en) * 2012-04-24 2013-11-14 Stmicroelectronics S.R.I. Multiplierless coprocessor for difference of gaussian (dog) calculation
CN103886589A (zh) * 2014-02-27 2014-06-25 四川农业大学 面向目标的自动化高精度边缘提取方法

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104268853A (zh) * 2014-03-06 2015-01-07 上海大学 一种红外图像与可见光图像配准方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101714254A (zh) * 2009-11-16 2010-05-26 哈尔滨工业大学 联合多尺度sift和区域不变矩特征的配准控制点提取方法
CN102646264A (zh) * 2011-02-18 2012-08-22 北京卫金帆医学技术发展有限公司 补偿机械运动误差的图像配准方法
CN102208109A (zh) * 2011-06-23 2011-10-05 南京林业大学 X射线图像和激光图像的异源图像配准方法
US20130301950A1 (en) * 2012-04-24 2013-11-14 Stmicroelectronics S.R.I. Multiplierless coprocessor for difference of gaussian (dog) calculation
CN103886589A (zh) * 2014-02-27 2014-06-25 四川农业大学 面向目标的自动化高精度边缘提取方法

Also Published As

Publication number Publication date
CN105761236A (zh) 2016-07-13

Similar Documents

Publication Publication Date Title
WO2017084261A1 (zh) 一种用于图像配准的图像预处理方法及装置
US20160217577A1 (en) Enhanced phase correlation for image registration
WO2019105044A1 (zh) 一种镜头畸变矫正和特征提取的方法及系统
JP6007602B2 (ja) 画像処理方法、画像処理装置、スキャナ及びコンピュータプログラム
US8867858B2 (en) Method and system for generating an output image of increased pixel resolution from an input image
US9224189B2 (en) Method and apparatus for combining panoramic image
US9396583B2 (en) Method of modelling buildings on the basis of a georeferenced image
US9959600B2 (en) Motion image compensation method and device, display device
CN104200461B (zh) 基于互信息图像选块和sift特征的遥感图像配准方法
KR20140109439A (ko) 잡음에 강한 영상 정합 방법 및 시스템
CN108765476B (zh) 一种偏振图像配准方法
Getreuer Contour stencils: Total variation along curves for adaptive image interpolation
JP2007520762A (ja) 画像位置合わせ方法の改良
CN110472521B (zh) 一种瞳孔定位校准方法及系统
Chan et al. Superresolution enhancement of hyperspectral CHRIS/Proba images with a thin-plate spline nonrigid transform model
Liu et al. Automatic blur-kernel-size estimation for motion deblurring
CN107240130A (zh) 遥感影像配准方法、装置及系统
Narváez et al. Point cloud denoising using robust principal component analysis
CN114119367A (zh) 一种分区域的同步相移干涉图超分辨重建的插值方法
US20170140538A1 (en) Image preprocessing method and electronic device for image registration
Shao et al. Edge-and-corner preserving regularization for image interpolation and reconstruction
JP2012203731A (ja) 画像の正置からの回転角度検知装置および方法
CN108537810B (zh) 一种改进的Zernike矩亚像素边缘检测方法
Chen et al. A wavelet transform based method for road centerline extraction
CN108647680A (zh) 图像定位框检测方法和装置

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: 16865443

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: 16865443

Country of ref document: EP

Kind code of ref document: A1