WO2019010932A1 - 一种利于模糊核估计的图像区域选择方法和系统 - Google Patents

一种利于模糊核估计的图像区域选择方法和系统 Download PDF

Info

Publication number
WO2019010932A1
WO2019010932A1 PCT/CN2018/071692 CN2018071692W WO2019010932A1 WO 2019010932 A1 WO2019010932 A1 WO 2019010932A1 CN 2018071692 W CN2018071692 W CN 2018071692W WO 2019010932 A1 WO2019010932 A1 WO 2019010932A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
pixel point
rtv
total variation
relative total
Prior art date
Application number
PCT/CN2018/071692
Other languages
English (en)
French (fr)
Inventor
桑农
李乐仁瀚
燕昊
高常鑫
颜露新
张士伟
皮智雄
Original Assignee
华中科技大学
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 华中科技大学 filed Critical 华中科技大学
Priority to US16/079,565 priority Critical patent/US11093778B2/en
Publication of WO2019010932A1 publication Critical patent/WO2019010932A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00Image enhancement or restoration
    • G06T5/73Deblurring; Sharpening
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00Arrangements for image or video recognition or understanding
    • G06V10/20Image preprocessing
    • G06V10/22Image preprocessing by selection of a specific region containing or referencing a pattern; Locating or processing of specific regions to guide the detection or recognition
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00Image analysis
    • G06T7/10Segmentation; Edge detection
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00Arrangements for image or video recognition or understanding
    • G06V10/40Extraction of image or video features
    • G06V10/44Local 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
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00Arrangements for image or video recognition or understanding
    • G06V10/40Extraction of image or video features
    • G06V10/50Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis

Definitions

  • the invention belongs to the field of image processing and pattern recognition, and more particularly to an image region selection method and system for facilitating fuzzy kernel estimation.
  • Image blurring is a common phenomenon of image degradation, which is caused by many reasons, such as: camera and shooting scene relative motion (motion blur), turbulence (turbulence blur) in the air due to high temperature, etc. There is a difference between the scene distance and the clear imaging focal length (defocus blur). Blurred images not only produce a poor viewing experience visually, but also affect the accuracy of computer vision tasks (such as image classification, target tracking, etc.) that require imagery. Therefore, the correction of blurred images (also called image deblurring) technology has become a key in the field of image processing and pattern recognition.
  • the process of image blurring is usually modeled as a clear image with a fuzzy kernel (also called a point spread function) for two-dimensional linear convolution.
  • a fuzzy kernel also called a point spread function
  • the purpose of image deblurring is to estimate a potentially sharp image based on the acquired blurred image.
  • Current image deblurring techniques are typically based on a least squares estimate that maximizes the posterior probability, and a "two-stage method" is used to estimate a sharp image.
  • the "two-stage method” first uses the relevant information in the image to estimate the fuzzy kernel by blind deconvolution, and then uses the estimated fuzzy kernel to estimate the clear image by non-blind deconvolution.
  • the large-scale and strong boundary regions in the image contain rich fuzzy kernel information, which is beneficial to the estimation of fuzzy kernels.
  • the present invention provides an image region selection method and system for blur kernel estimation, thereby solving the prior art experience of image region selection relying on operation, no scientific basis, and low efficiency. technical problem.
  • an image region selection method for blur kernel estimation including:
  • RTV(p)
  • RTV x (p) represents the relative total variation value of the pixel point p in the horizontal direction
  • RTV y (p) represents the relative total variation value of the pixel point p in the vertical direction
  • the relative total variation value of the pixel point p in the horizontal direction is:
  • R(p) represents a neighborhood centered on pixel p
  • q represents a pixel in the neighborhood. It represents the partial derivative of the pixel point q in the horizontal direction
  • is an infinitesimal quantity, ensuring that the denominator of the above formula is not zero
  • g p,q is a weight function, and its value is the distance between the pixel point q and the pixel point p.
  • inverse proportion
  • the relative total variation of the pixel point p in the vertical direction is:
  • step (3) is:
  • the blurred image B and its relative total variation map B rtv are superimposed pixel-sampled, and a sliding window is used to intercept an image block B i and a mapped image block for every s pixels.
  • An image block set P B ⁇ B 1 , B 2 , . . . , B i ⁇ and a set of mapped image blocks can be obtained by intercepting the image blocks from left to right and from top to bottom.
  • an image region selection system that facilitates fuzzy kernel estimation, including:
  • Relative total variational module used to calculate the relative total variation value RTV(p) of each pixel p in the blurred image B to obtain a relative total variation map B rtv which is the same as the blurred image size;
  • a determining module configured to determine that the pixel point p is a boundary pixel point when the relative total variation value RTV(p) of the pixel point p is less than a threshold; otherwise, determining that the pixel point p is a non-boundary pixel point;
  • a sampling module for sampling the blurred image B and its relative total variation map B rtv to obtain an image block B i and a mapped image block
  • an image block set P B ⁇ B 1 , B 2 , . . . , B i ⁇ and a set of mapped image blocks are obtained.
  • Area selection module for counting each mapped image block The number of pixels in the middle boundary, and in the collection Find the mapped image block with the largest number of boundary pixels
  • Corresponding image block An image area that is useful for blurring kernel estimation.
  • RTV(p)
  • RTV x (p) represents the relative total variation value of the pixel point p in the horizontal direction
  • RTV y (p) represents the relative total variation value of the pixel point p in the vertical direction
  • the relative total variation value of the pixel point p in the horizontal direction is:
  • R(p) represents a neighborhood centered on pixel p
  • q represents a pixel in the neighborhood. It represents the partial derivative of the pixel point q in the horizontal direction
  • is an infinitesimal quantity, ensuring that the denominator of the above formula is not zero
  • g p,q is a weight function, and its value is the distance between the pixel point q and the pixel point p.
  • inverse proportion
  • the relative total variation of the pixel point p in the vertical direction is:
  • sampling module is:
  • the blurred image B and its relative total variation map B rtv are superimposed pixel-sampled, and a sliding window is used to intercept an image block B i and a mapped image block for every s pixels.
  • An image block set P B ⁇ B 1 , B 2 , . . . , B i ⁇ and a set of mapped image blocks can be obtained by intercepting the image blocks from left to right and from top to bottom.
  • the present invention selects an image region which is advantageous for the fuzzy kernel estimation by measuring the relative total variation, and can improve the accuracy of the estimation result.
  • the present invention can automatically select an image region that is advantageous for fuzzy kernel estimation, and overcomes the disadvantage that the existing method relies on the operator's experience and is inefficient.
  • the present invention is particularly suitable for input image selection of fuzzy kernel estimation in an image deblurring algorithm.
  • FIG. 1 is a flowchart of an image region selection method for blur kernel estimation according to an embodiment of the present invention
  • FIG. 2(a) is a blurred image and its true blur kernel
  • FIG. 2(b) is an image result of blur kernel estimation and restoration using a full image
  • FIG. 2(c) is an image selected according to an embodiment of the present invention. The image results from the region where the fuzzy kernel is estimated and restored.
  • FIG. 1 it is a general flowchart of an image region selection method for facilitating fuzzy kernel estimation according to the present invention.
  • the method of the present invention specifically includes the following steps:
  • step (2) we set the threshold threshold to 0.1, and if the relative total variation value RTV(p) of the pixel point p is less than 0.1, it is determined to be a boundary pixel point; otherwise, it is non- Boundary pixel
  • the sliding pixel s is set to 20, and the size of the image block B i is 200 ⁇ 200.
  • the relative total variation value RTV(p) is:
  • RTV(p)
  • RTV x (p) represents the relative total variation value of the pixel point p in the horizontal direction
  • RTV y (p) represents the relative total variation value of the pixel point p in the vertical direction
  • R(p) represents a neighborhood centered on pixel p
  • q represents a pixel in the neighborhood. It represents the partial derivative of the pixel point q in the horizontal direction
  • is an infinitesimal quantity, ensuring that the denominator of the above formula is not zero
  • g p,q is a weight function, and its value is the distance between the pixel point q and the pixel point p.
  • inverse proportion
  • the relative total variation of the pixel point p in the vertical direction is:
  • x p is the abscissa of the pixel point p
  • y p is the ordinate of the pixel point p
  • x q is the abscissa of the pixel point q
  • y q is the ordinate of the pixel point q
  • exp( ⁇ ) is an exponential function
  • is the standard deviation.
  • FIG. 2 is a schematic diagram of image region selection results for blur kernel estimation
  • FIG. 2(a) is a blurred image and its true blur kernel
  • FIG. 2(b) is an image result of blur kernel estimation and restoration using a full image
  • 2(c) is an image result in which an image region is selected for blur kernel estimation and restored using the method proposed by the present invention. It can be seen from FIG. 2 that by performing the blurring kernel estimation image region selection method in the present invention and selecting an image region for fuzzy kernel estimation, a more accurate fuzzy kernel estimation result using the full map estimation can be obtained, and the utilization result is obtained.
  • the estimated fuzzy kernel is also ideal for image restoration. Comparing the restoration structure in FIG.
  • the technical solution of the present invention can automatically select an image block with more boundary regions for fuzzy kernel estimation, obtain more accurate and ideal estimation results and restoration effects, and significantly improve the recovery efficiency of each algorithm. It is therefore especially suitable for the field of image deblurring.
  • the invention adopts the measurement method of relative total variation to select the structural region rich in "large scale and strong boundary" in the blurred image as the input of the fuzzy kernel estimation phase in the image deblurring algorithm, and improves the accuracy and operation of the fuzzy kernel estimation.
  • the efficiency effectively solves the problems of image region selection and operation experience, no scientific basis, and low efficiency existing in the existing methods.

Landscapes

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

Abstract

一种利于模糊核估计的图像区域选择方法和系统,其中方法的实现包括:计算模糊图像中每一像素点的相对总变分值并得到其相对总变分映射图;设定阈值确定图像中每一像素点是否为边界像素点;再对模糊图像以及其相对总变分映射图进行采样,得到一系列图像块;最后统计每一图像块中边界像素点的数量并选择出有利于模糊核估计的图像区域。该方法有效解决了现有区域选择方法中存在的过于依赖操作者经验,效率低等问题,自动选择出有利于模糊核估计的图像区域,适用于图像去模糊算法中模糊核估计的应用场合。

Description

一种利于模糊核估计的图像区域选择方法和系统 [技术领域]
本发明属于图像处理、模式识别领域,更具体地,涉及一种利于模糊核估计的图像区域选择方法和系统。
[背景技术]
图像模糊是一种常见的图像退化现象,其产生的原因有很多,如:相机与拍摄场景在曝光时发生相对运动(运动模糊)、空气中因高温等原因出现湍流(湍流模糊)以及相机与场景距离和清晰成像焦距存在差异(失焦模糊)等。模糊的图像不仅在视觉上产生较差的观赏体验,而且在一些需要利用图像进行的计算机视觉任务(如图像分类、目标跟踪等)中影响其准确性。因此,将模糊图像进行校正(也称图像去模糊)技术已成为图像处理、模式识别领域中的一大关键。
图像模糊的过程通常被建模为一个清晰图像与模糊核(也称点扩展函数)进行二维线性卷积。图像去模糊的目的是根据已获得的模糊图像估计出其潜在的清晰图像。目前的图像去模糊技术通常基于最大化后验概率的最小二乘估计,并且采用“两阶段法”估计清晰图像。“两阶段法”首先利用图像中的相关信息采用盲反卷积估计出模糊核,再利用估计出的模糊核采用非盲反卷积估计出清晰图像。图像中的大尺度、强边界区域包含较为丰富的模糊核信息,有利于模糊核的估计;而图像中的平滑、纹理区域对模糊核估计无帮助、甚至会影响模糊核估计的准确度。因此,选择一个有利于模糊核估计的图像区域至关重要。对于图像区域的选取方法主要包括三类:(1)全图输入,该方法直接将全图作为输入对模糊核进行估计,当图中平滑区域、纹理区域较多时,模糊核估计结果往往不准确,并且全图输入会使计算量增大;(2)基于经验选取结构区域较多的图像块,该方 法通常依赖于操作者的经验,通过“试错法”选取图像区域,无科学依据且效率较低。(3)自动选择,采用的机器学习的方法,此过程包含训练与推断两个部分,训练部分需要大量的数据且耗时较长。
由此可见,现有技术存在图像区域选择依赖操作经验、无科学依据、效率低的技术问题。
[发明内容]
针对现有技术的以上缺陷或改进需求,本发明提供了一种利于模糊核估计的图像区域选择方法和系统,由此解决现有技术存在图像区域选择依赖操作经验、无科学依据、效率低的技术问题。
为实现上述目的,按照本发明的一个方面,提供了一种利于模糊核估计的图像区域选择方法,包括:
(1)计算模糊图像B中每一像素点p的相对总变分值RTV(p)得到与模糊图像尺寸相同的相对总变分映射图B rtv
(2)当像素点p的相对总变分值RTV(p)小于阈值时,判定像素点p为边界像素点;否则,判定像素点p为非边界像素点;
(3)对模糊图像B及其相对总变分映射图B rtv进行采样,得到图像块B i和映射图像块
Figure PCTCN2018071692-appb-000001
截取图像块后得到一个图像块集合P B={B 1,B 2,...,B i}以及映射图像块的集合
Figure PCTCN2018071692-appb-000002
(4)统计每个映射图像块
Figure PCTCN2018071692-appb-000003
中边界像素点的个数,并在集合
Figure PCTCN2018071692-appb-000004
中找到含有边界像素点个数最多的映射图像块
Figure PCTCN2018071692-appb-000005
Figure PCTCN2018071692-appb-000006
对应的图像块
Figure PCTCN2018071692-appb-000007
为利于模糊核估计的图像区域。
进一步的,相对总变分值RTV(p)为:
RTV(p)=|RTV x(p)|+|RTV y(p)|
Figure PCTCN2018071692-appb-000008
其中,RTV x(p)表示像素点p在水平方向上的相对总变分值,RTV y(p)表示像素点p在竖直方向上的相对总变分值。
进一步的,像素点p在水平方向上的相对总变分值为:
Figure PCTCN2018071692-appb-000009
其中,R(p)表示以像素点p为中心的邻域,q表示邻域中的像素点,
Figure PCTCN2018071692-appb-000010
表示像素点q在水平方向上的偏导数,ε为一无穷小量,保证上式分母不为零,g p,q为一权重函数,其取值与像素点q和像素点p之间的距离成反比;
像素点p在竖直方向上的相对总变分值为:
Figure PCTCN2018071692-appb-000011
其中,
Figure PCTCN2018071692-appb-000012
表示像素点q在竖直方向上的偏导数。
进一步的,步骤(3)的具体实现方式为:
对模糊图像B及其相对总变分映射图B rtv进行重叠像素式的采样,采用滑动窗口的方式,每滑动s个像素,截取一个尺寸为m×m图像块B i及映射图像块
Figure PCTCN2018071692-appb-000013
从左至右、从上至下截取图像块后可得到一个图像块集合P B={B 1,B 2,...,B i}以及映射图像块的集合
Figure PCTCN2018071692-appb-000014
按照本发明的另一方面,提供了一种利于模糊核估计的图像区域选择系统,包括:
相对总变分模块,用于计算模糊图像B中每一像素点p的相对总变分值RTV(p)得到与模糊图像尺寸相同的相对总变分映射图B rtv
判定模块,用于当像素点p的相对总变分值RTV(p)小于阈值时,判定像素点p为边界像素点;否则,判定像素点p为非边界像素点;
采样模块,用于对模糊图像B及其相对总变分映射图B rtv进行采样,得到图像块B i和映射图像块
Figure PCTCN2018071692-appb-000015
截取图像块后得到一个图像块集合P B={B 1,B 2,...,B i}以及映射图像块的集合
Figure PCTCN2018071692-appb-000016
区域选择模块,用于统计每个映射图像块
Figure PCTCN2018071692-appb-000017
中边界像素点的个数,并在集合
Figure PCTCN2018071692-appb-000018
中找到含有边界像素点个数最多的映射图像块
Figure PCTCN2018071692-appb-000019
Figure PCTCN2018071692-appb-000020
对应的图像块
Figure PCTCN2018071692-appb-000021
为利于模糊核估计的图像区域。
进一步的,相对总变分值RTV(p)为:
RTV(p)=|RTV x(p)|+|RTV y(p)|
Figure PCTCN2018071692-appb-000022
其中,RTV x(p)表示像素点p在水平方向上的相对总变分值,RTV y(p)表示像素点p在竖直方向上的相对总变分值。
进一步的,像素点p在水平方向上的相对总变分值为:
Figure PCTCN2018071692-appb-000023
其中,R(p)表示以像素点p为中心的邻域,q表示邻域中的像素点,
Figure PCTCN2018071692-appb-000024
表示像素点q在水平方向上的偏导数,ε为一无穷小量,保证上式分母不为零,g p,q为一权重函数,其取值与像素点q和像素点p之间的距离成反比;
像素点p在竖直方向上的相对总变分值为:
Figure PCTCN2018071692-appb-000025
其中,
Figure PCTCN2018071692-appb-000026
表示像素点q在竖直方向上的偏导数。
进一步的,采样模块的具体实现方式为:
对模糊图像B及其相对总变分映射图B rtv进行重叠像素式的采样,采用滑动窗口的方式,每滑动s个像素,截取一个尺寸为m×m图像块B i及映射图像块
Figure PCTCN2018071692-appb-000027
从左至右、从上至下截取图像块后可得到一个图像块集合P B={B 1,B 2,...,B i}以及映射图像块的集合
Figure PCTCN2018071692-appb-000028
总体而言,通过本发明所构思的以上技术方案与现有技术相比,能够取得下列有益效果:
(1)本发明通过相对总变分的度量方式选取有利于模糊核估计的图像区域,能提升其估计结果的准确性。
(2)本发明可以自动地选择有利于模糊核估计的图像区域,克服了现有方法依赖操作者经验,效率较低的缺点。
(3)本发明中通过相对总变分的度量方式确定像素点是否为边界像素点,并通过统计图像区域中的边界像素点判定其是否为有利于模糊核估计的图像区域,方法简洁清楚,因此本发明尤其适用于图像去模糊算法中模糊核估计的输入图像选取。
[附图说明]
图1是本发明实施例提供的一种利于模糊核估计的图像区域选择方法的流程图;
图2(a)为模糊图像以及其真实的模糊核,图2(b)为利用全图进行的模糊核估计并复原的图像结果,图2(c)为本发明实施例提供的选择一个图像区域进行模糊核估计并复原的图像结果。
[具体实施方式]
为了使本发明的目的、技术方案及优点更加清楚明白,以下结合附图及实施例,对本发明进行进一步详细说明。应当理解,此处所描述的具体实施例仅仅用以解释本发明,并不用于限定本发明。此外,下面所描述的本发明各个实施方式中所涉及到的技术特征只要彼此之间未构成冲突就可以相互组合。
如图1所示,为本发明利于模糊核估计的图像区域选择方法总流程图,本发明方法具体包括以下步骤:
(1)计算模糊图像B中每一像素点p的相对总变分值RTV(p)得到与模糊图像尺寸相同的相对总变分映射图B rtv
(2)设定一阈值threshold,当像素点p的相对总变分值RTV(p)小于阈值时,则判定其为边界像素点;反之,则为非边界像素点;
(3)对模糊图像B及其相对总变分映射图B rtv进行重叠像素式的采样,采用滑动窗口的方式,每滑动s个像素,截取一个尺寸为m×m图像块B i及映射图像块
Figure PCTCN2018071692-appb-000029
从左至右、从上至下截取图像块后可得到一个图像块集合P B={B 1,B 2,...,B i}以及映射图像块的集合
Figure PCTCN2018071692-appb-000030
(4)统计每个映射图像块
Figure PCTCN2018071692-appb-000031
中边界像素点的个数,并在集合
Figure PCTCN2018071692-appb-000032
中找到含有边界像素点个数最多的映射图像块
Figure PCTCN2018071692-appb-000033
其对应的图像块
Figure PCTCN2018071692-appb-000034
则为本方法选择出的利于模糊核估计的图像区域。
优选地,在步骤(2)中,我们将阈值threshold设定为0.1,若像素点p的相对总变分值RTV(p)小于0.1时,则判定其为边界像素点;反之,则为非边界像素点;
优选地,在步骤(3)中,设定滑动像素s为20,图像块B i的尺寸为200×200。
优选的,相对总变分值RTV(p)为:
RTV(p)=|RTV x(p)|+|RTV y(p)|
Figure PCTCN2018071692-appb-000035
其中,RTV x(p)表示像素点p在水平方向上的相对总变分值,RTV y(p)表示像素点p在竖直方向上的相对总变分值。
像素点p在水平方向上的相对总变分值为:
Figure PCTCN2018071692-appb-000036
其中,R(p)表示以像素点p为中心的邻域,q表示邻域中的像素点,
Figure PCTCN2018071692-appb-000037
表示像素点q在水平方向上的偏导数,ε为一无穷小量,保证上式分母不为零,g p,q为一权重函数,其取值与像素点q和像素点p之间的距离成反比;
像素点p在竖直方向上的相对总变分值为:
Figure PCTCN2018071692-appb-000038
其中,
Figure PCTCN2018071692-appb-000039
表示像素点q在竖直方向上的偏导数。
Figure PCTCN2018071692-appb-000040
式中,x p为像素点p的横坐标,y p为像素点p的纵坐标,x q为像素点q的横坐标,y q为像素点q的纵坐标,exp(□)为指数函数,σ为标准差。
图2为利于模糊核估计的图像区域选择结果示意图,图2(a)为模糊图像以及其真实的模糊核,图2(b)为利用全图进行的模糊核估计并复原的图像结果,图2(c)为利用本发明提出的方法,选择一个图像区域进行模糊核估计并复原的图像结果。从图2可以看出,通过执行本发明中的利 于模糊核估计图像区域选择方法,选择一个图像区域进行模糊核估计,可得到较利用全图估计的更准确的模糊核估计结果,并且利用所估计出的模糊核进行图像复原的结果也较理想。对比图2中的复原结构,可知本发明技术方案可自动选择出一个具有较多边界区域的图像块进行模糊核估计,得到较为准确理想的估计结果以及复原效果,显著提升了各算法复原效率,因而尤其适用于图像去模糊领域。
本发明采用相对总变分的度量方法选择出模糊图像中富含“大尺度、强边界”的结构区域作为图像去模糊算法中模糊核估计阶段的输入,提升了模糊核估计的准确性以及操作效率,有效解决了现有方法中存在的图像区域选择依赖操作经验、无科学依据、效率低等问题。
本领域的技术人员容易理解,以上所述仅为本发明的较佳实施例而已,并不用以限制本发明,凡在本发明的精神和原则之内所作的任何修改、等同替换和改进等,均应包含在本发明的保护范围之内。

Claims (8)

  1. 一种利于模糊核估计的图像区域选择方法,其特征在于,包括:
    (1)计算模糊图像B中每一像素点p的相对总变分值RTV(p)得到与模糊图像尺寸相同的相对总变分映射图B rtv
    (2)当像素点p的相对总变分值RTV(p)小于阈值时,判定像素点p为边界像素点;否则,判定像素点p为非边界像素点;
    (3)对模糊图像B及其相对总变分映射图B rtv进行采样,得到图像块B i和映射图像块
    Figure PCTCN2018071692-appb-100001
    截取图像块后得到一个图像块集合P B={B 1,B 2,…,B i}以及映射图像块的集合
    Figure PCTCN2018071692-appb-100002
    (4)统计每个映射图像块
    Figure PCTCN2018071692-appb-100003
    中边界像素点的个数,并在集合
    Figure PCTCN2018071692-appb-100004
    中找到含有边界像素点个数最多的映射图像块
    Figure PCTCN2018071692-appb-100005
    Figure PCTCN2018071692-appb-100006
    对应的图像块
    Figure PCTCN2018071692-appb-100007
    为利于模糊核估计的图像区域。
  2. 如权利要求1所述的一种利于模糊核估计的图像区域选择方法,其特征在于,所述相对总变分值RTV(p)为:
    RTV(p)=|RTV x(p)|+|RTV y(p)|
    Figure PCTCN2018071692-appb-100008
    其中,RTV x(p)表示像素点p在水平方向上的相对总变分值,RTV y(p)表示像素点p在竖直方向上的相对总变分值。
  3. 如权利要求2所述的一种利于模糊核估计的图像区域选择方法,其特征在于,所述像素点p在水平方向上的相对总变分值为:
    Figure PCTCN2018071692-appb-100009
    其中,R(p)表示以像素点p为中心的邻域,q表示邻域中的像素点,
    Figure PCTCN2018071692-appb-100010
    表示像素点q在水平方向上的偏导数,ε为一无穷小量,保证上式分母不为零,g p,q为一权重函数,其取值与像素点q和像素点p之间的距离成反比;
    所述像素点p在竖直方向上的相对总变分值为:
    Figure PCTCN2018071692-appb-100011
    其中,
    Figure PCTCN2018071692-appb-100012
    表示像素点q在竖直方向上的偏导数。
  4. 如权利要求1或2或3所述的一种利于模糊核估计的图像区域选择方法,其特征在于,所述步骤(3)的具体实现方式为:
    对模糊图像B及其相对总变分映射图B rtv进行重叠像素式的采样,采用滑动窗口的方式,每滑动s个像素,截取一个尺寸为m×m图像块B i及映射图像块
    Figure PCTCN2018071692-appb-100013
    从左至右、从上至下截取图像块后可得到一个图像块集合P B={B 1,B 2,…,B i}以及映射图像块的集合
    Figure PCTCN2018071692-appb-100014
  5. 一种利于模糊核估计的图像区域选择系统,其特征在于,包括:
    相对总变分模块,用于计算模糊图像B中每一像素点p的相对总变分值RTV(p)得到与模糊图像尺寸相同的相对总变分映射图B rtv
    判定模块,用于当像素点p的相对总变分值RTV(p)小于阈值时,判定像素点p为边界像素点;否则,判定像素点p为非边界像素点;
    采样模块,用于对模糊图像B及其相对总变分映射图B rtv进行采样,得到图像块B i和映射图像块
    Figure PCTCN2018071692-appb-100015
    截取图像块后得到一个图像块集合 P B={B 1,B 2,…,B i}以及映射图像块的集合
    Figure PCTCN2018071692-appb-100016
    区域选择模块,用于统计每个映射图像块
    Figure PCTCN2018071692-appb-100017
    中边界像素点的个数,并在集合
    Figure PCTCN2018071692-appb-100018
    中找到含有边界像素点个数最多的映射图像块
    Figure PCTCN2018071692-appb-100019
    对应的图像块
    Figure PCTCN2018071692-appb-100020
    为利于模糊核估计的图像区域。
  6. 如权利要求5所述的一种利于模糊核估计的图像区域选择系统,其特征在于,所述相对总变分值RTV(p)为:
    RTV(p)=|RTV x(p)|+|RTV y(p)|
    Figure PCTCN2018071692-appb-100021
    其中,RTV x(p)表示像素点p在水平方向上的相对总变分值,RTV y(p)表示像素点p在竖直方向上的相对总变分值。
  7. 如权利要求6所述的一种利于模糊核估计的图像区域选择系统,其特征在于,所述像素点p在水平方向上的相对总变分值为:
    Figure PCTCN2018071692-appb-100022
    其中,R(p)表示以像素点p为中心的邻域,q表示邻域中的像素点,
    Figure PCTCN2018071692-appb-100023
    表示像素点q在水平方向上的偏导数,ε为一无穷小量,保证上式分母不为零,g p,q为一权重函数,其取值与像素点q和像素点p之间的距离成反比;
    所述像素点p在竖直方向上的相对总变分值为:
    Figure PCTCN2018071692-appb-100024
    其中,
    Figure PCTCN2018071692-appb-100025
    表示像素点q在竖直方向上的偏导数。
  8. 如权利要求5或6或7所述的一种利于模糊核估计的图像区域选择系统,其特征在于,所述采样模块的具体实现方式为:
    对模糊图像B及其相对总变分映射图B rtv进行重叠像素式的采样,采用滑动窗口的方式,每滑动s个像素,截取一个尺寸为m×m图像块B i及映射图像块
    Figure PCTCN2018071692-appb-100026
    从左至右、从上至下截取图像块后可得到一个图像块集合P B={B 1,B 2,…,B i}以及映射图像块的集合
    Figure PCTCN2018071692-appb-100027
PCT/CN2018/071692 2017-07-14 2018-01-08 一种利于模糊核估计的图像区域选择方法和系统 WO2019010932A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US16/079,565 US11093778B2 (en) 2017-07-14 2018-01-08 Method and system for selecting image region that facilitates blur kernel estimation

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201710572664.9 2017-07-14
CN201710572664.9A CN107403414B (zh) 2017-07-14 2017-07-14 一种利于模糊核估计的图像区域选择方法和系统

Publications (1)

Publication Number Publication Date
WO2019010932A1 true WO2019010932A1 (zh) 2019-01-17

Family

ID=60405412

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/071692 WO2019010932A1 (zh) 2017-07-14 2018-01-08 一种利于模糊核估计的图像区域选择方法和系统

Country Status (3)

Country Link
US (1) US11093778B2 (zh)
CN (1) CN107403414B (zh)
WO (1) WO2019010932A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109859145A (zh) * 2019-02-27 2019-06-07 长安大学 一种基于多级权重相对总变差的图像去纹理方法
CN113362362A (zh) * 2021-06-17 2021-09-07 易普森智慧健康科技(深圳)有限公司 基于总变分区域选取的明视野显微镜全景图像对齐算法

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107403414B (zh) 2017-07-14 2018-11-02 华中科技大学 一种利于模糊核估计的图像区域选择方法和系统
US11615510B2 (en) * 2020-10-21 2023-03-28 Samsung Electronics Co., Ltd. Kernel-aware super resolution
US20230041888A1 (en) * 2021-08-03 2023-02-09 Samsung Electronics Co., Ltd. Unsupervised learning-based scale-independent blur kernel estimation for super-resolution
CN113870134B (zh) * 2021-09-28 2023-04-07 南京诺源医疗器械有限公司 手持式摄像镜数据处理方法及装置

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102254309A (zh) * 2011-07-27 2011-11-23 清华大学 一种基于近红外图像的运动模糊图像去模糊方法和装置
CN102436639A (zh) * 2011-09-02 2012-05-02 清华大学 一种去除图像模糊的图像采集方法和图像采集系统
CN102663718A (zh) * 2012-03-19 2012-09-12 清华大学 一种全局不一致图像去模糊的方法及系统
CN105096316A (zh) * 2015-06-30 2015-11-25 华为技术有限公司 一种确定图像模糊程度的方法及装置
CN107403414A (zh) * 2017-07-14 2017-11-28 华中科技大学 一种利于模糊核估计的图像区域选择方法和系统

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9659351B2 (en) * 2014-03-12 2017-05-23 Purdue Research Foundation Displaying personalized imagery for improving visual acuity
CN104168336A (zh) 2014-09-05 2014-11-26 北京奇虎科技有限公司 数据获取方法、移动设备、计算设备及系统
US10325358B2 (en) * 2015-05-15 2019-06-18 Peking University Shenzhen Graduate School Method and system for image de-blurring

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102254309A (zh) * 2011-07-27 2011-11-23 清华大学 一种基于近红外图像的运动模糊图像去模糊方法和装置
CN102436639A (zh) * 2011-09-02 2012-05-02 清华大学 一种去除图像模糊的图像采集方法和图像采集系统
CN102663718A (zh) * 2012-03-19 2012-09-12 清华大学 一种全局不一致图像去模糊的方法及系统
CN105096316A (zh) * 2015-06-30 2015-11-25 华为技术有限公司 一种确定图像模糊程度的方法及装置
CN107403414A (zh) * 2017-07-14 2017-11-28 华中科技大学 一种利于模糊核估计的图像区域选择方法和系统

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109859145A (zh) * 2019-02-27 2019-06-07 长安大学 一种基于多级权重相对总变差的图像去纹理方法
CN109859145B (zh) * 2019-02-27 2022-12-09 西安汇智信息科技有限公司 一种基于多级权重相对总变差的图像去纹理方法
CN113362362A (zh) * 2021-06-17 2021-09-07 易普森智慧健康科技(深圳)有限公司 基于总变分区域选取的明视野显微镜全景图像对齐算法

Also Published As

Publication number Publication date
US11093778B2 (en) 2021-08-17
CN107403414B (zh) 2018-11-02
CN107403414A (zh) 2017-11-28
US20210192251A1 (en) 2021-06-24

Similar Documents

Publication Publication Date Title
WO2019010932A1 (zh) 一种利于模糊核估计的图像区域选择方法和系统
CN106875415B (zh) 一种动态背景中弱小动目标的连续稳定跟踪方法
CN107256225B (zh) 一种基于视频分析的热度图生成方法及装置
WO2022141178A1 (zh) 图像处理方法及装置
CN106886748B (zh) 一种基于tld的适用于无人机的变尺度目标跟踪方法
US9589328B2 (en) Globally dominant point spread function estimation
CN107316326B (zh) 应用于双目立体视觉的基于边的视差图计算方法和装置
CN108470356B (zh) 一种基于双目视觉的目标对象快速测距方法
CN109974743B (zh) 一种基于gms特征匹配及滑动窗口位姿图优化的视觉里程计
CN109657607B (zh) 一种基于人脸识别的人脸目标测距方法、装置及存储介质
CN107516322B (zh) 一种基于对数极空间的图像物体大小和旋转估计计算方法
WO2017012418A1 (zh) 图像处理方法及装置
CN111340749B (zh) 图像质量的检测方法、装置、设备及存储介质
CN110345875B (zh) 标定及测距方法、装置、电子设备及计算机可读存储介质
CN112946679B (zh) 一种基于人工智能的无人机测绘果冻效应检测方法及系统
US11741615B2 (en) Map segmentation method and device, motion estimation method, and device terminal
CN109035170A (zh) 基于单网格图分段映射的自适应广角图像校正方法及装置
WO2021057455A1 (zh) 用于红外图像序列的背景运动估计方法、装置及存储介质
CN114494085B (zh) 一种视频流复原方法、系统、电子设备及存储介质
JP2019096222A (ja) 画像処理装置、画像処理方法、コンピュータプログラム
CN112991159B (zh) 人脸光照质量评估方法、系统、服务器与计算机可读介质
JP7312026B2 (ja) 画像処理装置、画像処理方法およびプログラム
CN106296688B (zh) 基于全局估计的影像模糊检测方法及系统
CN109741370B (zh) 一种目标跟踪方法和装置
CN115345845A (zh) 基于方向梯度直方图的特征融合烟幕干扰效能评估及处理方法及电子设备

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: 18831593

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: 18831593

Country of ref document: EP

Kind code of ref document: A1