WO2019127059A1 - 基于引导滤波和高通滤波的红外图像非均匀性校正方法 - Google Patents

基于引导滤波和高通滤波的红外图像非均匀性校正方法 Download PDF

Info

Publication number
WO2019127059A1
WO2019127059A1 PCT/CN2017/118768 CN2017118768W WO2019127059A1 WO 2019127059 A1 WO2019127059 A1 WO 2019127059A1 CN 2017118768 W CN2017118768 W CN 2017118768W WO 2019127059 A1 WO2019127059 A1 WO 2019127059A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
frame image
uniformity
filtering
current frame
Prior art date
Application number
PCT/CN2017/118768
Other languages
English (en)
French (fr)
Inventor
周慧鑫
赵东
钱润达
贾秀萍
周峻
秦翰林
姚博
于跃
李欢
宋江鲁奇
王炳健
金阳群
荣生辉
成宽洪
钱琨
Original Assignee
西安电子科技大学
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 西安电子科技大学 filed Critical 西安电子科技大学
Priority to PCT/CN2017/118768 priority Critical patent/WO2019127059A1/zh
Priority to US16/338,703 priority patent/US10803557B2/en
Publication of WO2019127059A1 publication Critical patent/WO2019127059A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/70—Denoising; Smoothing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/50—Image enhancement or restoration using two or more images, e.g. averaging or subtraction
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/20—Image enhancement or restoration using local operators
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/20—Analysis of motion
    • G06T7/254—Analysis of motion involving subtraction of images
    • 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
    • G06T2207/10—Image acquisition modality
    • G06T2207/10016—Video; Image sequence
    • 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
    • G06T2207/10—Image acquisition modality
    • G06T2207/10048—Infrared image
    • 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
    • G06T2207/20172—Image enhancement details
    • G06T2207/20182—Noise reduction or smoothing in the temporal domain; Spatio-temporal filtering
    • 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
    • G06T2207/20212—Image combination
    • G06T2207/20224—Image subtraction
    • 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
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30232—Surveillance

Definitions

  • the invention belongs to the field of infrared image processing, and in particular relates to a method for correcting infrared image non-uniformity based on guided filtering and high-pass filtering.
  • Infrared focal plane array is the core component of modern infrared imaging system. Due to its small size, light weight and no need for refrigeration equipment, it is widely used in military reconnaissance, forest fire prevention and medical detection; however, infrared coke Planar arrays, due to the limitations of their own materials and manufacturing processes, produce different pattern noise on the infrared image even if different pixels in the same radiation incident produce different outputs, ie response non-uniformity. , seriously affecting the quality of imaging.
  • the non-uniformity correction methods of infrared images are mainly divided into two major categories: calibration methods and scene methods.
  • the calibration method needs to periodically pause the work to re-calibrate to obtain new correction parameters, and the real-time performance is poor. Therefore, the scene method has become the main research object in recent years; the scene method estimates the scene information and relies on the motion of the scene.
  • the correction parameters are updated, and the correction parameters can be adaptively updated; typical scene methods include time domain high-pass filtering, neural network, constant statistics, Kalman filtering, and inter-frame registration.
  • the scene method takes a long time because of its complicated algorithm, and makes a wrong estimation of the correction parameters when the stationary target suddenly starts moving and the moving target suddenly stops, thereby generating a "ghosting" phenomenon.
  • the main object of the present invention is to provide an infrared image non-uniformity correction method based on guided filtering and high-pass filtering.
  • Embodiments of the present invention provide a method for correcting infrared image non-uniformity based on guided filtering and high-pass filtering, which is: performing guided filtering on an input first image of an original image sequence with non-uniformity to obtain an original image sequence.
  • the non-uniformity correction is performed on the current frame image according to the nth fixed pattern noise f n to obtain a correction result of the current frame image, and then the method further includes: determining whether the current frame image is The last frame of the original image sequence with non-uniformity, if yes, the correction is completed; if not, the subsequent frame image is continued for correction.
  • the first frame image of the input original image sequence with non-uniformity is guided and filtered to obtain a high frequency component of the first frame image of the original image sequence, and the high frequency component of the first frame image of the original image sequence is obtained.
  • the component assignment is the first fixed pattern noise f 1 , which is achieved by the following steps:
  • the Nth frame image in the original image sequence with non-uniformity is sequentially loaded as the current frame image, and the current frame image and the original image sequence with non-uniformity are determined to be N-1.
  • the difference of the frame image is obtained by the difference image of the N-1th frame, which is specifically implemented by the following steps:
  • the relative change amplitude of each pixel of the N-1th frame image is obtained according to the N-1th frame difference image, specifically: determining the N-1th frame image according to the following formula: The relative magnitude of each pixel:
  • q n-1 (i, j) represents the relative variation amplitude of the i-th row and the j-th column of the image of the N-1th frame
  • d n-1 (i, j) represents the N-1th difference image
  • the gray value of the j-th column pixel of the i-th row, x n-1 (i, j) represents the gray value of the i-th column and the j-th column of the image of the N-1th frame.
  • the high-frequency component of the current frame image obtained by the guided filtering and the relative variation amplitude of each pixel of the N-1th frame image are high-pass filtered to obtain the nth fixed pattern noise f n , specifically:
  • the nth fixed pattern noise f n is determined according to the following formula:
  • f n (i, j) represents the gray value of the i-th row and j-th column of the nth fixed pattern noise
  • M min represents a small time constant of the time domain high-pass filtering
  • M max represents the time domain high-pass filtering Large time constant
  • f n-1 (i, j) represents the gray of the i-th row and the j-th column of the n-1th fixed pattern noise
  • q n-1 (i, j) represents the relative variation amplitude of the i-th row j-th column pixel of the N-1th frame image
  • Th represents the threshold value, 0.1 ⁇ Th ⁇ 0.3.
  • the frame image is subjected to non-uniformity correction to obtain a current frame image correction result; wherein y n represents the current frame image correction result, I n represents the current frame image, and f n represents the nth (n ⁇ 2) fixed pattern noise.
  • the filtering process is performed on the first frame image of the original image sequence with non-uniformity according to the guiding filtering, and the current frame image is filtered according to the guiding filtering, which is specifically implemented by the following steps:
  • w k represents to the k-th pixel window centered
  • is the total number of cells in the w k
  • z denotes an image number w k of elements
  • I z represents the original image z-th picture elements Gray value
  • p z is the gray value of the zth pixel of the guided image
  • u k and ⁇ k are the mean and standard deviation of the original image I in w k , respectively.
  • ⁇ is a positive number with a small value, which is 0.01 in the present invention
  • the high-frequency components of the image can be effectively separated, which overcomes the high residual non-uniformity caused by the inaccuracy and incomplete separation of the high-frequency components in the prior art.
  • the ghosting problem enables the present invention to effectively perform subsequent motion-based time-domain high-pass filtering.
  • the motion difference between the two adjacent frames is determined by the inter-frame difference method to find the motion region and the still region. Then, in the time domain high-pass filtering, a smaller time constant is used for the corresponding motion region of the difference image.
  • the corresponding still region is filtered by a large time constant, and the processed difference image contains less edge and detail information, overcoming the incomplete removal of the fixed pattern noise and the edge caused by the fixed time constant in the prior art. Problems such as blurring enable the present invention to preserve the edge and detail information of the image while removing the fixed pattern noise.
  • Figure 1 is a flow chart of the present invention
  • FIG. 2 is a first frame image of an original image sequence with non-uniformity input according to the present invention
  • 3 is a low frequency component of the first frame image of the original image sequence with non-uniformity after the guided filtering process in the present invention
  • Figure 5 is a 700th frame image of an original image sequence with non-uniformity in the present invention.
  • Figure 9 is a final correction result of the 700th frame image of the original image sequence with non-uniformity in the present invention.
  • Embodiments of the present invention provide a method for correcting infrared image non-uniformity based on boot filtering and high-pass filtering. As shown in FIG. 1 , the method is:
  • Step 1 reading in the first frame image of the original image sequence with non-uniformity
  • FIG. 2 is a first frame image of an original image sequence with non-uniformity in an embodiment of the present invention
  • the original image sequence with non-uniformity has a total of 700 frames, and the image size of each frame is 292. ⁇ 200 pixels; as can be seen from Figure 2, the original image has a noticeable fixed pattern noise.
  • Step 2 Filtering the first frame image of the original image sequence with non-uniformity according to the booting filter to obtain a low frequency component of the first frame image of the original image sequence;
  • the specific steps of the boot filtering are as follows:
  • Step 201 Calculate the pilot filter multiplicative parameter a k and the boot filter additive parameter b k according to the following formula:
  • w k represents to the k-th pixel window centered
  • is the total number of cells in the w k
  • z denotes an image number w k of elements
  • I z represents the original image z-th picture elements Gray value
  • p z is the gray value of the zth pixel of the guided image
  • u k and ⁇ k are the mean and standard deviation of the original image I in w k , respectively.
  • ⁇ is a positive number with a small value, which is 0.02 in the present invention.
  • Step 202 Calculate the mean value of the guided filter multiplicative parameter according to the following formula And guiding filter additive parameter mean
  • Step 203 Calculate the guided filtered image x L according to the following formula:
  • FIG. 3 is a low-frequency component of a first frame image of an original image sequence with non-uniformity according to an embodiment of the present invention.
  • the image belongs to a high-frequency component.
  • the noise is suppressed while retaining the edge information of the image, but there are blurring effects on the details and edges of the person in the image.
  • Step 3 subtracting the low frequency component of the first frame image of the original image sequence by using the first frame image of the original image sequence with non-uniformity to obtain the high frequency component of the first frame image of the original image sequence;
  • FIG. 4 is a high-frequency component of the first frame image of the original image sequence with non-uniformity after being guided and filtered in the embodiment of the present invention. As can be seen from FIG. 4, it contains a large amount of detailed information and edge.
  • Step 4 assigning a high frequency component of the first frame image of the original image sequence to the first fixed pattern noise f 1 ;
  • Step 5 Loading the Nth frame image in the original image sequence with non-uniformity as the current frame image
  • FIG. 5 is a 700th frame image of a sequence of original images with non-uniformity in an embodiment of the present invention.
  • Step 6 Filtering the current frame image by using the boot filter to obtain a low frequency component of the current frame image
  • the specific steps of the boot filtering are as follows.
  • Step 601 calculating a pilot filter parameter multiplicative parameter a k and a boot filter additive parameter b k according to the following formula:
  • w k represents to the k-th pixel window centered
  • is the total number of cells in the w k
  • z denotes an image number w k of elements
  • I z represents the original image z-th picture elements Gray value
  • p z is the gray value of the zth pixel of the guided image
  • u k and ⁇ k are the mean and standard deviation of the original image I in w k , respectively.
  • ⁇ is a positive number with a small value, which is 0.02 in the present invention.
  • Step 602 calculating a mean value of the guided filter multiplicative parameter according to the following formula And guiding filter additive parameter mean
  • Step 603 calculating the guided filtered image x L according to the following formula:
  • FIG. 6 is a low frequency component of the 700th frame image of the original image sequence with non-uniformity after being guided and filtered in the embodiment of the present invention
  • Step 7 subtracting the low frequency component of the current frame image by using the current frame image to obtain a high frequency component of the current frame image
  • FIG. 7 is a high frequency component of the 700th frame image of the original image sequence with non-uniformity after being guided and filtered in the embodiment of the present invention
  • Step 8 calculating a difference between the current frame image and the N-1th frame image to obtain an N-1th frame difference image
  • Step 9 Calculate the relative variation amplitude of each pixel in the N-1th frame image according to the following formula:
  • q n-1 (i, j) represents the relative variation amplitude of the i-th row and the j-th column of the image of the N-1th frame
  • d n-1 (i, j) represents the N-1th difference image
  • the gray value of the j-th column pixel of the i-th row, x n-1 (i, j) represents the gray value of the i-th column and the j-th column of the image of the N-1th frame;
  • Step 10 Calculate the nth fixed pattern noise f n according to the following formula:
  • f n (i, j) represents the gray value of the i-th row and j-th column of the nth fixed pattern noise
  • M min represents a small time constant of the time-domain high-pass filtering, which takes a value of 2
  • M max is expressed Large time constant of domain high-pass filtering, value 5
  • f n-1 (i, j) represents the gray of the i-th row and the j-th column of the n-1th fixed pattern noise
  • q n-1 (i, j) represents the relative variation amplitude of the i-th row and the j-th column of the image of the N-1th frame
  • Th represents the threshold, 0.1 ⁇ Th ⁇ 0.3;
  • FIG. 8 is a fixed pattern noise of a 700th frame image of an original image sequence with non-uniformity in an embodiment of the present invention.
  • Step 11 Perform non-uniformity correction on the current frame image according to the following formula to obtain a current frame image correction result:
  • FIG. 9 is a result of correction of the 700th frame image of the original image sequence with non-uniformity in the embodiment of the present invention.
  • Step 12 determining whether the current frame image is the last frame image of the original image sequence with non-uniformity, and if so, performing step (13); otherwise, performing step (5);
  • Step 13 Complete the non-uniformity correction of the original image sequence with non-uniformity.
  • the infrared image sequence non-uniformity correction method based on the guidance filtering and the high-pass filtering proposed by the invention firstly obtains the low-frequency component of the original image by guiding filtering, and then subtracts the low-frequency component of the original image from the original image to obtain the high-frequency component of the original image. Then the inter-frame difference method is used to judge the motion region and the still region in the original image, and then different time constants are set for the motion region and the still region to perform high-pass filtering on the high-frequency components of the original image to obtain a fixed boundary. Pattern noise, eventually subtracting the fixed pattern noise from the original image to achieve non-uniformity correction.

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)
  • Transforming Light Signals Into Electric Signals (AREA)

Abstract

本发明公开了一种基于引导滤波和高通滤波的红外图像非均匀性校正方法，将带有非均匀性的原始图像序列第一帧图像的高频成分赋值为第一个固定图案噪声f 1；依次载入带有非均匀性的原始图像序列中第N帧图像作为当前帧图像，确定当前帧图像和第N-1帧图像的差值获得第N-1帧差值图像，根据第N-1帧差值图像获得所述第N-1帧图像每个像元的相对变化幅度；结合当前帧图像的高频成分和第N-1帧差值图像每个像元的变化幅度进行高通滤波获得第n个固定图案噪声f n，根据第n个固定图案噪声f n对当前帧图像进行非均匀性校正，获得当前帧图像的校正结果；其中，N≥2，n≥2。本发明能够实现对由非均匀性引起的红外图像序列的固定图案噪声进行有效地去除。

Description

基于引导滤波和高通滤波的红外图像非均匀性校正方法 技术领域
本发明属于红外图像处理领域，具体涉及一种基于引导滤波和高通滤波的红外图像非均匀性校正方法。
背景技术
红外焦平面阵列是现代红外成像系统中的核心部件，由于其具有体积小、重量轻，不需要制冷设备等优点而被广泛地应用于军事侦察、森林防火和医疗检测等领域；然而，红外焦平面阵列由于自身材料和制作工艺的水平的限制，即使在相同的辐射入射的条件下不同像元也会产生不同输出，即响应非均匀性，响应非均匀性在红外图像上产生了固定图案噪声，严重影响成像质量。
目前，红外图像的非均匀性校正方法主要分为定标法和场景法两大类。定标法需要周期性地暂停工作来重新定标以获取新的校正参数，实时性较差，所以近年来场景法成为了主要的研究对象；场景法对场景信息进行估计，依靠场景的运动来更新校正参数，并且可以对校正参数进行自适应更新；典型的场景法包括时域高通滤波法、神经网络法、恒定统计法、卡尔曼滤波法以及帧间配准法。
场景法由于其算法复杂因而所需的时间较长，并且在静止目标突然开始运动和运动目标突然静止的情况下会对校正参数作出错误的估计，进而产生“鬼影”现象。
发明内容
有鉴于此，本发明的主要目的在于提供一种基于引导滤波和高通滤波的红外图像非均匀性校正方法。
为达到上述目的，本发明的技术方案是这样实现的：
本发明实施例提供一种基于引导滤波和高通滤波的红外图像非均匀性校正方法，该方法为：对输入的带有非均匀性的原始图像序列第一帧图像进行引导 滤波获得原始图像序列第一帧图像的高频成分，并且将原始图像序列第一帧图像的高频成分赋值为第一个固定图案噪声f 1；依次载入所述带有非均匀性的原始图像序列中第N帧图像作为当前帧图像，确定所述当前帧图像和带有非均匀性的原始图像序列第N-1帧图像的差值获得第N-1帧差值图像,根据所述第N-1帧差值图像获得所述第N-1帧图像每个像元的相对变化幅度；结合经过引导滤波获得的当前帧图像的高频成分和第N-1帧图像每个像元的相对变化幅度进行高通滤波获得第n个固定图案噪声f n，根据所述第n个固定图案噪声f n对当前帧图像进行非均匀性校正，获得当前帧图像的校正结果；其中，N≥2，n≥2。
上述方案中，所述根据所述第n个固定图案噪声f n对当前帧图像进行非均匀性校正，获得当前帧图像的校正结果，之后，该方法还包括，判断所述当前帧图像是否为带有非均匀性的原始图像序列的最后一帧图像，如果是则完成校正；如果不是则继续后续帧图像进行校正。
上述方案中，所述对输入的带有非均匀性的原始图像序列第一帧图像进行引导滤波获得原始图像序列第一帧图像的高频成分，并且将原始图像序列第一帧图像的高频成分赋值为第一个固定图案噪声f 1，具体通过以下步骤实现：
(101)读入带有非均匀性的原始图像序列第一帧图像；
(102)根据引导滤波对带有非均匀性的原始图像序列第一帧图像进行滤波处理，得到原始图像序列第一帧图像的低频成分；
(103)根据带有非均匀性的原始图像序列第一帧图像减去原始图像序列第一帧图像的低频成分，得到原始图像序列第一帧图像的高频成分；
(104)将原始图像序列第一帧图像的高频成分赋值为第一个固定图案噪声f 1。
上述方案中，所述依次载入所述带有非均匀性的原始图像序列中第N帧图像作为当前帧图像，确定所述当前帧图像和带有非均匀性的原始图像序列第N-1帧图像的差值获得第N-1帧差值图像，具体通过以下步骤实现：
(201)依次载入带有非均匀性的原始图像序列中第N帧图像作为当前帧图像；
(202)根据引导滤波对当前帧图像进行滤波处理，得到当前帧图像的低频成分；
(203)根据当前帧图像减去当前帧图像的低频成分，得到当前帧图像的高频成分；
(204)计算当前帧图像和第N-1帧图像的差值，得到第N-1帧差值图像。
上述方案中，所述根据所述第N-1帧差值图像获得所述第N-1帧图像每个像元的相对变化幅度，具体为：根据下式确定所述第N-1帧图像每个像元的相对变化幅度：
Figure PCTCN2017118768-appb-000001
其中，q n-1(i,j)表示第N-1帧图像第i行第j列像元的相对变化幅度，d n-1(i,j)表示第N-1帧差值图像第i行第j列像元的灰度值，x n-1(i,j)表示第N-1帧图像第i行第j列像元的灰度值。
上述方案中，所述结合经过引导滤波获得的当前帧图像的高频成分和第N-1帧图像每个像元的相对变化幅度进行高通滤波获得第n个固定图案噪声f n，具体为：根据下式确定第n个固定图案噪声f n：
Figure PCTCN2017118768-appb-000002
其中，f n(i,j)表示第n个固定图案噪声第i行第j列像元的灰度值，M min表示时域高通滤波的较小时间常数，M max表示时域高通滤波的较大时间常数，
Figure PCTCN2017118768-appb-000003
表示当前帧图像的高频成分第i行第j列像元的灰度值，f n-1(i,j)表示第n-1个固定图案噪声第i行第j列像元的灰度值，q n-1(i,j)表示第N-1帧图像第i行第j列像元的相对变化幅度，Th表示阈值，0.1≤Th≤0.3。
上述方案中，所述根据所述第n个固定图案噪声f n对当前帧图像进行非均匀性校正，获得当前帧图像的校正结果，具体为：根据公式y n＝I n-f n对当前帧图像进行非均匀性校正，得到当前帧图像校正结果；其中，y n表示当前帧图像校 正结果，I n表示当前帧图像，f n表示第n(n≥2)个固定图案噪声。
上述方案中，所述根据引导滤波对带有非均匀性的原始图像序列第一帧图像进行滤波处理和根据引导滤波对当前帧图像进行滤波处理，具体通过如下步骤实现：
(301)根据下式确定引导滤波乘性参数a k和引导滤波加性参数b k：
Figure PCTCN2017118768-appb-000004
Figure PCTCN2017118768-appb-000005
其中，w k表示以第k个像元为中心的窗口，|w|为w k内的像元总数，z表示w k中的像元的序号，I z表示原始图像第z个像元的灰度值，p z为引导图像第z个像元的灰度值，u k和σ k分别为原始图像I在w k内的均值和标准差，
Figure PCTCN2017118768-appb-000006
为p在w k内的均值，ε为一个数值很小的正数，本发明中取0.01；
(302)根据下式确定引导滤波乘性参数均值
Figure PCTCN2017118768-appb-000007
和引导滤波加性参数均值
Figure PCTCN2017118768-appb-000008
Figure PCTCN2017118768-appb-000009
Figure PCTCN2017118768-appb-000010
(303)按照下式确定引导滤波后的图像x L：
Figure PCTCN2017118768-appb-000011
与现有技术相比，本发明的有益效果：
第一、由于本发明的引导滤波算法，能有效地分离出图像的高频成分，克服了现有技术由于对高频成分分离得不准确和不彻底造成的剩余非均匀性较高以及产生“鬼影”问题，使得本发明能够有效地进行后续的基于运动判断的时域高通滤波。
第二、采用帧间差分法对相邻两帧原始图像进行运动判断，找出运动区域 和静止区域，然后在时域高通滤波中对差值图像相应的运动区域使用较小的时间常数，对相应的静止区域使用较大的时间常数进行滤波，经过处理后的差值图像含有较少的边缘和细节信息，克服了现有技术中采用固定的时间常数造成的固定图案噪声去除不彻底以及边缘模糊等问题，使得本发明能够在去除固定图案噪声的同时很好地保留图像的边缘和细节信息。
附图说明
图1为本发明的流程图；
图2为本发明输入带有非均匀性的原始图像序列第一帧图像；
图3为本发明中带有非均匀性的原始图像序列第一帧图像经引导滤波处理后的低频成分；
图4为本发明中带有非均匀性的原始图像序列第一帧图像经引导滤波处理后的高频成分；
图5为本发明中带有非均匀性的原始图像序列的第700帧图像；
图6为本发明中带有非均匀性的原始图像序列第700帧图像经引导滤波处理后的低频成分；
图7为本发明中带有非均匀性的原始图像序列第700帧图像经引导滤波处理后的高频成分；
图8为本发明中经基于运动判断的时域高通滤波法得到的带有非均匀性的原始图像序列的第700帧图像的固定图案噪声；
图9为本发明中带有非均匀性的原始图像序列的第700帧图像的最终校正结果。
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。
本发明实施例提供一种基于引导滤波和高通滤波的红外图像非均匀性校正方法，如图1所示该方法为：
步骤1：读入带有非均匀性的原始图像序列第一帧图像；
具体地，图2为本发明的实施例中，带有非均匀性的原始图像序列第一帧图像；带有非均匀性的原始图像序列共有700帧图像，并且每一帧图像大小都为292×200像素；从图2可以看出，原始图像带有明显的固定图案噪声。
步骤2：根据引导滤波对带有非均匀性的原始图像序列第一帧图像进行滤波处理，得到原始图像序列第一帧图像的低频成分；
具体地，所述引导滤波的具体步骤如下：
步骤201，按照下式计算引导滤波乘性参数a k和引导滤波加性参数b k：
Figure PCTCN2017118768-appb-000012
Figure PCTCN2017118768-appb-000013
其中，w k表示以第k个像元为中心的窗口，|w|为w k内的像元总数，z表示w k中的像元的序号，I z表示原始图像第z个像元的灰度值，p z为引导图像第z个像元的灰度值，u k和σ k分别为原始图像I在w k内的均值和标准差，
Figure PCTCN2017118768-appb-000014
为p在w k内的均值，ε为一个数值很小的正数，本发明中取0.02。
步骤202，按照下式计算引导滤波乘性参数均值
Figure PCTCN2017118768-appb-000015
和引导滤波加性参数均值
Figure PCTCN2017118768-appb-000016
Figure PCTCN2017118768-appb-000017
Figure PCTCN2017118768-appb-000018
步骤203，按照下式计算引导滤波后的图像x L：
Figure PCTCN2017118768-appb-000019
具体地，图3为本发明的实施例中，带有非均匀性的原始图像序列第一帧图像的低频成分，从图3可以看出，由于使用引导滤波算法，图像中属于高频成分的噪声得到抑制，同时保留了图像的边缘信息，但是图像中人身上的细节 和边缘存在模糊效应。
步骤3：利用带有非均匀性的原始图像序列第一帧图像减去原始图像序列第一帧图像的低频成分，得到原始图像序列第一帧图像的高频成分；
具体地，图4为本发明的实施例中，带有非均匀性的原始图像序列第一帧图像经引导滤波后的高频成分，从图4可以看出，其包含了大量的细节信息和边缘。
步骤4：将原始图像序列第一帧图像的高频成分赋值给第一个固定图案噪声f 1；
步骤5：载入带有非均匀性的原始图像序列中第N帧图像，作为当前帧图像；
具体地，图5为本发明的实施例中，带有非均匀性的原始图像序列中第700帧图像。
步骤6：使用引导滤波对当前帧图像进行滤波处理，得到当前帧图像的低频成分；
具体地，所述引导滤波的具体步骤如下。
步骤601，按照下式计算引导滤波参数乘性参数a k和引导滤波加性参数b k：
Figure PCTCN2017118768-appb-000020
Figure PCTCN2017118768-appb-000021
其中，w k表示以第k个像元为中心的窗口，|w|为w k内的像元总数，z表示w k中的像元的序号，I z表示原始图像第z个像元的灰度值，p z为引导图像第z个像元的灰度值，u k和σ k分别为原始图像I在w k内的均值和标准差，
Figure PCTCN2017118768-appb-000022
为p在w k内的均值，ε为一个数值很小的正数，本发明中取0.02。
步骤602，按照下式计算引导滤波乘性参数均值
Figure PCTCN2017118768-appb-000023
和引导滤波加性参数均值
Figure PCTCN2017118768-appb-000024
Figure PCTCN2017118768-appb-000025
Figure PCTCN2017118768-appb-000026
步骤603，按照下式计算引导滤波后的图像x L：
Figure PCTCN2017118768-appb-000027
具体地，图6为本发明的实施例中，带有非均匀性的原始图像序列第700帧图像经引导滤波处理后的低频成分；
步骤7：利用当前帧图像减去当前帧图像的低频成分，得到当前帧图像的高频成分；
具体地，图7为本发明的实施例中，带有非均匀性的原始图像序列第700帧图像经引导滤波处理后的高频成分；
步骤8：计算当前帧图像和第N-1帧图像的差值，得到第N-1帧差值图像；
步骤9：按照下式计算第N-1帧图像每个像元的相对变化幅度：
Figure PCTCN2017118768-appb-000028
其中，q n-1(i,j)表示第N-1帧图像第i行第j列像元的相对变化幅度，d n-1(i,j)表示第N-1帧差值图像第i行第j列像元的灰度值，x n-1(i,j)表示第N-1帧图像第i行第j列像元的灰度值；
步骤10：按照下式计算第n个固定图案噪声f n：
Figure PCTCN2017118768-appb-000029
其中，f n(i,j)表示第n个固定图案噪声第i行第j列像元的灰度值，M min表示时域高通滤波的较小时间常数，取值2，M max表示时域高通滤波的较大时间常数，取值5，
Figure PCTCN2017118768-appb-000030
表示当前帧图像的高频成分第i行第j列像元的灰度值，f n-1(i,j)表示第n-1个固定图案噪声第i行第j列像元的灰度值，q n-1(i,j)表示第 N-1帧图像第i行第j列像元的相对变化幅度，Th表示阈值，0.1≤Th≤0.3；
具体地，图8为本发明的实施例中，带有非均匀性的原始图像序列第700帧图像的固定图案噪声。
步骤11：按照下式，对当前帧图像进行非均匀性校正，得到当前帧图像校正结果：
y n＝I n-f n
其中，y n表示当前帧图像校正结果，I n表示当前帧图像，f n表示第n个固定图案噪声；
具体地，图9为本发明的实施例中，带有非均匀性的原始图像序列第700帧图像的校正结果。
步骤12：判断当前帧图像是否为带有非均匀性的原始图像序列的最后一帧图像，若是，则执行步骤(13)；否则，执行步骤(5)；
步骤13：完成带有非均匀性的原始图像序列的非均匀性校正。
本发明提出的基于引导滤波和高通滤波的红外图像序列非均匀性校正方法首先通过引导滤波得到原始图像的低频成分，然后从原始图像中减去原始图像的低频成分得到原始图像的高频成分，再通过帧间差分法判断原始图像中的运动区域和静止区域，再对运动区域和静止区域设置不同的时间常数来对原始图像的高频成分进行时域高通滤波，得到几乎不含有边界的固定图案噪声，最终从原始图像中减去固定图案噪声，实现非均匀性校正。
以上所述，仅为本发明的较佳实施例而已，并非用于限定本发明的保护范围。

Claims (8)

  1. 一种基于引导滤波和高通滤波的红外图像非均匀性校正方法，其特征在于，该方法为：对输入的带有非均匀性的原始图像序列第一帧图像进行引导滤波获得原始图像序列第一帧图像的高频成分，并且将原始图像序列第一帧图像的高频成分赋值为第一个固定图案噪声f 1；依次载入所述带有非均匀性的原始图像序列中第N帧图像作为当前帧图像，确定所述当前帧图像和带有非均匀性的原始图像序列第N-1帧图像的差值获得第N-1帧差值图像,根据所述第N-1帧差值图像获得所述第N-1帧图像每个像元的相对变化幅度；结合经过引导滤波获得的当前帧图像的高频成分和第N-1帧图像每个像元的相对变化幅度进行高通滤波获得第n个固定图案噪声f n，根据所述第n个固定图案噪声f n对当前帧图像进行非均匀性校正，获得当前帧图像的校正结果；其中，N≥2，n≥2。
  2. 根据权利要求1所述的基于引导滤波和高通滤波的红外图像非均匀性校正方法，其特征在于，所述根据所述第n个固定图案噪声f n对当前帧图像进行非均匀性校正，获得当前帧图像的校正结果，之后，该方法还包括，判断所述当前帧图像是否为带有非均匀性的原始图像序列的最后一帧图像，如果是则完成校正；如果不是则继续后续帧图像进行校正。
  3. 根据权利要求1或2所述的一种基于引导滤波和高通滤波的红外图像非均匀性校正方法，其特征在于，所述对输入的带有非均匀性的原始图像序列第一帧图像进行引导滤波获得原始图像序列第一帧图像的高频成分，并且将原始图像序列第一帧图像的高频成分赋值为第一个固定图案噪声f 1，具体通过以下步骤实现：
    (101)读入带有非均匀性的原始图像序列第一帧图像；
    (102)根据引导滤波对带有非均匀性的原始图像序列第一帧图像进行滤波处理，得到原始图像序列第一帧图像的低频成分；
    (103)根据带有非均匀性的原始图像序列第一帧图像减去原始图像序列第一帧图像的低频成分，得到原始图像序列第一帧图像的高频成分；
    (104)将原始图像序列第一帧图像的高频成分赋值为第一个固定图案噪声f 1。
  4. 根据权利要求3所述的一种基于引导滤波和高通滤波的红外图像非均匀性校正方法，其特征在于，所述依次载入所述带有非均匀性的原始图像序列中第N帧图像作为当前帧图像，确定所述当前帧图像和带有非均匀性的原始图像序列第N-1帧图像的差值获得第N-1帧差值图像，具体通过以下步骤实现：
    (201)依次载入带有非均匀性的原始图像序列中第N帧图像作为当前帧图像；
    (202)根据引导滤波对当前帧图像进行滤波处理，得到当前帧图像的低频成分；
    (203)根据当前帧图像减去当前帧图像的低频成分，得到当前帧图像的高频成分；
    (204)计算当前帧图像和第N-1帧图像的差值，得到第N-1帧差值图像。
  5. 根据权利要求4所述的一种基于引导滤波和高通滤波的红外图像非均匀性校正方法，其特征在于，所述根据所述第N-1帧差值图像获得所述第N-1帧图像每个像元的相对变化幅度，具体为：根据下式确定所述第N-1帧图像每个像元的相对变化幅度：
    Figure PCTCN2017118768-appb-100001
    其中，q n-1(i,j)表示第N-1帧图像第i行第j列像元的相对变化幅度，d n-1(i,j)表示第N-1帧差值图像第i行第j列像元的灰度值，x n-1(i,j)表示第N-1帧图像第i行第j列像元的灰度值。
  6. 根据权利要求5所述的一种基于引导滤波和高通滤波的红外图像非均匀性校正方法，其特征在于，所述结合经过引导滤波获得的当前帧图像的高频成分和第N-1帧图像每个像元的相对变化幅度进行高通滤波获得第n个固定图案噪声f n，具体为：根据下式确定第n个固定图案噪声f n：
    Figure PCTCN2017118768-appb-100002
    其中，f n(i,j)表示第n个固定图案噪声第i行第j列像元的灰度值，M min表示时域高通滤波的较小时间常数，M max表示时域高通滤波的较大时间常数，
    Figure PCTCN2017118768-appb-100003
    表示当前帧图像的高频成分第i行第j列像元的灰度值，f n-1(i,j)表示第n-1个固定图案噪声第i行第j列像元的灰度值，q n-1(i,j)表示第N-1帧图像第i行第j列像元的相对变化幅度，Th表示阈值，0.1≤Th≤0.3。
  7. 根据权利要求6所述的一种基于引导滤波和高通滤波的红外图像非均匀性校正方法，其特征在于，所述根据所述第n个固定图案噪声f n对当前帧图像进行非均匀性校正，获得当前帧图像的校正结果，具体为：根据公式y n＝I n-f n对当前帧图像进行非均匀性校正，得到当前帧图像校正结果；其中，y n表示当前帧图像校正结果，I n表示当前帧图像，f n表示第n(n≥2)个固定图案噪声。
  8. 根据权利要求7所述的一种基于引导滤波和高通滤波的红外图像非均匀性校正方法，其特征在于，所述根据引导滤波对带有非均匀性的原始图像序列第一帧图像进行滤波处理和根据引导滤波对当前帧图像进行滤波处理，具体通过如下步骤实现：
    (301)根据下式确定引导滤波乘性参数a k和引导滤波加性参数b k：
    Figure PCTCN2017118768-appb-100004
    Figure PCTCN2017118768-appb-100005
    其中，w k表示以第k个像元为中心的窗口，|w|为w k内的像元总数，z表示w k中的像元的序号，I z表示原始图像第z个像元的灰度值，p z为引导图像第z个像元的灰度值，u k和σ k分别为原始图像I在w k内的均值和标准差，
    Figure PCTCN2017118768-appb-100006
    为p在w k内的均值，ε为一个数值很小的正数，本发明中取0.01；
    (302)根据下式确定引导滤波乘性参数均值
    Figure PCTCN2017118768-appb-100007
    和引导滤波加性参数均值
    Figure PCTCN2017118768-appb-100008
    Figure PCTCN2017118768-appb-100009
    Figure PCTCN2017118768-appb-100010
    (303)按照下式确定引导滤波后的图像x L：
    Figure PCTCN2017118768-appb-100011
PCT/CN2017/118768 2017-12-26 2017-12-26 基于引导滤波和高通滤波的红外图像非均匀性校正方法 WO2019127059A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
PCT/CN2017/118768 WO2019127059A1 (zh) 2017-12-26 2017-12-26 基于引导滤波和高通滤波的红外图像非均匀性校正方法
US16/338,703 US10803557B2 (en) 2017-12-26 2017-12-26 Non-uniformity correction method for infrared image based on guided filtering and high-pass filtering

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2017/118768 WO2019127059A1 (zh) 2017-12-26 2017-12-26 基于引导滤波和高通滤波的红外图像非均匀性校正方法

Publications (1)

Publication Number Publication Date
WO2019127059A1 true WO2019127059A1 (zh) 2019-07-04

Family

ID=67062778

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/118768 WO2019127059A1 (zh) 2017-12-26 2017-12-26 基于引导滤波和高通滤波的红外图像非均匀性校正方法

Country Status (2)

Country Link
US (1) US10803557B2 (zh)
WO (1) WO2019127059A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110796621A (zh) * 2019-10-29 2020-02-14 浙江大华技术股份有限公司 红外图像去横纹处理方法、处理设备和存储装置
CN111383196A (zh) * 2020-03-13 2020-07-07 浙江大华技术股份有限公司 红外图像条纹消除方法、红外探测器及存储装置

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP7147507B2 (ja) * 2018-11-22 2022-10-05 コニカミノルタ株式会社 画像処理装置及びプログラム
CN112258479B (zh) * 2020-10-22 2023-10-27 中国人民解放军63620部队 基于图像特征的t0检测方法、装置和存储介质
US11544918B2 (en) 2020-12-30 2023-01-03 Adasky, Ltd. Vehicle to infrastructure system and method with long wave infrared capability
CN113379636B (zh) * 2021-06-21 2024-05-03 苏州睿新微系统技术有限公司 一种红外图像非均匀性校正方法、装置、设备及存储介质

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101776486A (zh) * 2009-12-31 2010-07-14 华中科技大学 一种基于红外焦平面非均匀性指纹模式的校正方法
CN105005967A (zh) * 2015-05-28 2015-10-28 西安电子科技大学 时空滤波相结合的红外成像非均匀性校正方法及其装置
CN105160657A (zh) * 2015-08-05 2015-12-16 西安电子科技大学 基于fpga的红外成像非均匀性校正系统
CN105890768A (zh) * 2016-03-31 2016-08-24 浙江大华技术股份有限公司 一种红外图像非均匀性校正的方法及装置
CN106153198A (zh) * 2015-04-22 2016-11-23 南京理工大学 一种基于时域高通滤波的帧间配准非均匀性校正方法
US20170358067A1 (en) * 2014-12-02 2017-12-14 University Of Seoul Industry Cooperation Foundation Method and Device for Fusing Panchromatic Image and Infrared Image

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP2760330A4 (en) * 2011-09-30 2015-04-01 Univ Columbia METHODS, AND COMPACT IMAGING SYSTEMS AND OPTICAL DEVICES
FR3020735B1 (fr) * 2014-04-30 2017-09-15 Ulis Procede de traitement d'une image infrarouge pour une correction des non uniformites
US9924116B2 (en) * 2014-08-05 2018-03-20 Seek Thermal, Inc. Time based offset correction for imaging systems and adaptive calibration control
US20190339688A1 (en) * 2016-05-09 2019-11-07 Strong Force Iot Portfolio 2016, Llc Methods and systems for data collection, learning, and streaming of machine signals for analytics and maintenance using the industrial internet of things
CN106780392B (zh) * 2016-12-27 2020-10-02 浙江大华技术股份有限公司 一种图像融合方法及装置

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101776486A (zh) * 2009-12-31 2010-07-14 华中科技大学 一种基于红外焦平面非均匀性指纹模式的校正方法
US20170358067A1 (en) * 2014-12-02 2017-12-14 University Of Seoul Industry Cooperation Foundation Method and Device for Fusing Panchromatic Image and Infrared Image
CN106153198A (zh) * 2015-04-22 2016-11-23 南京理工大学 一种基于时域高通滤波的帧间配准非均匀性校正方法
CN105005967A (zh) * 2015-05-28 2015-10-28 西安电子科技大学 时空滤波相结合的红外成像非均匀性校正方法及其装置
CN105160657A (zh) * 2015-08-05 2015-12-16 西安电子科技大学 基于fpga的红外成像非均匀性校正系统
CN105890768A (zh) * 2016-03-31 2016-08-24 浙江大华技术股份有限公司 一种红外图像非均匀性校正的方法及装置

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110796621A (zh) * 2019-10-29 2020-02-14 浙江大华技术股份有限公司 红外图像去横纹处理方法、处理设备和存储装置
CN110796621B (zh) * 2019-10-29 2022-06-07 浙江大华技术股份有限公司 红外图像去横纹处理方法、处理设备和存储装置
CN111383196A (zh) * 2020-03-13 2020-07-07 浙江大华技术股份有限公司 红外图像条纹消除方法、红外探测器及存储装置
CN111383196B (zh) * 2020-03-13 2023-07-28 浙江大华技术股份有限公司 红外图像条纹消除方法、红外探测器及存储装置

Also Published As

Publication number Publication date
US20200143517A1 (en) 2020-05-07
US10803557B2 (en) 2020-10-13

Similar Documents

Publication Publication Date Title
WO2019127059A1 (zh) 基于引导滤波和高通滤波的红外图像非均匀性校正方法
Abdelhamed et al. A high-quality denoising dataset for smartphone cameras
CN109272520B (zh) 一种联合运动指导与边缘检测的自适应红外焦平面非均匀校正方法
Wang et al. Enhancing low light videos by exploring high sensitivity camera noise
CN106373105B (zh) 基于低秩矩阵恢复的多曝光图像去伪影融合方法
CN108665423A (zh) 基于引导滤波和高通滤波的红外图像非均匀性校正方法
CN106197690B (zh) 一种宽温范围条件下的图像校准方法及系统
CN104796582B (zh) 基于随机喷射retinex的视频图像去噪与增强方法及装置
US9554058B2 (en) Method, apparatus, and system for generating high dynamic range image
CN106525245B (zh) 一种基于三梯度阈值的快速时序盲元检测与校正方法
CN109813442B (zh) 一种基于多帧处理的内部杂散辐射非均匀性校正方法
CN109934790A (zh) 带有自适应阈值的红外成像系统非均匀性校正方法
CN111598812B (zh) 一种基于rgb和hsv双颜色空间的图像去雾方法
WO2019183843A1 (zh) 基于帧间配准和自适应步长的红外图像非均匀性校正方法
JP2016540440A (ja) ピクチャー処理方法、装置
Li et al. Scene-based nonuniformity correction based on bilateral filter with reduced ghosting
CN111080561A (zh) 一种时域高通滤波方法
CN103997592B (zh) 视频降噪方法和系统
CN106651812A (zh) 一种简单镜头成像的多通道psf标定方法
CN105787892A (zh) 一种基于机器学习的蒙特卡洛噪声去除方法
CN111405206B (zh) 一种泊松-高斯联合噪声图像序列分离降噪方法
CN113432723B (zh) 用于弱化杂散辐射的图像处理方法、系统及计算机系统
CN105869129A (zh) 针对非均匀校正后的热红外图像剩余非均匀噪声去除方法
Li et al. Single-frame-based column fixed-pattern noise correction in an uncooled infrared imaging system based on weighted least squares
CN107610072B (zh) 一种基于梯度引导滤波的微光视频图像自适应降噪方法

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: 17936472

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: 17936472

Country of ref document: EP

Kind code of ref document: A1