WO2016169244A1 - 基于随机喷射retinex的视频图像去噪与增强方法及装置 - Google Patents

基于随机喷射retinex的视频图像去噪与增强方法及装置 Download PDF

Info

Publication number
WO2016169244A1
WO2016169244A1 PCT/CN2015/094088 CN2015094088W WO2016169244A1 WO 2016169244 A1 WO2016169244 A1 WO 2016169244A1 CN 2015094088 W CN2015094088 W CN 2015094088W WO 2016169244 A1 WO2016169244 A1 WO 2016169244A1
Authority
WO
WIPO (PCT)
Prior art keywords
brightness
video image
denoising
retinex
calculation formula
Prior art date
Legal status (The legal status 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 status listed.)
Ceased
Application number
PCT/CN2015/094088
Other languages
English (en)
French (fr)
Inventor
程德强
李雷达
刘海
张国鹏
郭星歌
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
China University of Mining and Technology Beijing CUMTB
Original Assignee
China University of Mining and Technology Beijing CUMTB
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 China University of Mining and Technology Beijing CUMTB filed Critical China University of Mining and Technology Beijing CUMTB
Priority to US15/305,350 priority Critical patent/US9978131B2/en
Publication of WO2016169244A1 publication Critical patent/WO2016169244A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/90—Dynamic range modification of images or parts thereof
    • G06T5/94—Dynamic range modification of images or parts thereof based on local image properties, e.g. for local contrast enhancement
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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 OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/70—Denoising; Smoothing
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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 OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10024—Color image
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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 OR CALCULATING; 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
    • G06T2207/20201—Motion blur correction
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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
    • G06T2207/20208—High dynamic range [HDR] image processing

Definitions

  • the invention relates to the field of image processing technologies, and in particular to a video image denoising and enhancement method and apparatus based on random injection retinex.
  • the low quality of video images is mainly affected by two aspects: (1)
  • the general video capture system can capture good quality images on sunny days, but in the case of bad weather such as fog or dust, or insufficient nighttime illumination. Under the system, the contrast of images captured by the system is often very low, and it is difficult for people to obtain valuable information from the screen.
  • Video image enhancement technology can effectively improve the visual effect of video images, highlight the information that people are interested in, and discard some useless information.
  • the video image In the process of acquisition or transmission, the video image usually degrades due to noise pollution, and the video image deviates from the real situation, which greatly affects the accuracy of people extracting information from it. Therefore, it is necessary to use video. Eliminate noise before the image, improve the quality of the image after noise cancellation, and highlight the characteristics of the video image.
  • Gray image enhancement method is more mature, with contrast enhancement method and histogram equalization Method, homomorphic filtering method, wavelet transform method, etc.
  • the color image enhancement method mainly has the method based on Retinex theory. This method has good characteristics in dynamic range compression and color constancy, such as retinex algorithm based on path comparison, but this algorithm has large computational complexity and geometry. The problem of high path dependence and the presence of sample noise does not yield good results for noisy images. Therefore, there is a need to improve prior art algorithms based on Retinex theory.
  • the technical solution of the present invention provides a video image denoising and enhancement method based on random injection retinex, comprising: constructing a set of jet pixel points, and performing parameters related to random injection retinex processing based on the set of injected pixel points.
  • a tuning process wherein the parameter comprises a number of injection sets and a number of pixel points; a random injection retinex process is performed on the video image based on the parameter after the tuning process; and a low-pass filter is performed on the video image of the random injection retinex process
  • Denoising processing and fuzzy channel denoising processing obtaining a brightness change calculation formula; combining the brightness change calculation formula to obtain a brightness calculation formula of the output image, and using the brightness calculation formula to calculate the brightness change values of the three channels to form a local brightness Estimating the vector; and based on the obtained local luminance estimation vector, performing three-channel fusion to obtain a de-noised and enhanced video image.
  • constructing the set of jetted pixel points in a polar coordinate manner comprises:
  • the number of pixel points centered on the pixel point i is represented by Spray k (i), and the coordinates of the pixel point i are (i x , i y ), and the coordinates of the pixel point j belonging to the ejection set Spray k (i) can be set as (j x ,j y ), the polar coordinates are:
  • ⁇ is an arbitrary value of the interval [0, R]
  • R represents the ejection radius
  • ⁇ is an arbitrary value of the interval [0, 2 ⁇ ].
  • the parameters involved in the random injection retinex process are tuned, wherein the parameters further include a spray radius and a radius density function.
  • the tuning process for the jet radius involved in the random injection retinex process comprises selecting the diagonal length of the video image to be the jet radius.
  • the tuning process for the number of ejection sets and the number of pixel points involved in the random injection retinex processing comprises: selecting the number of ejection sets to be 1, and selecting the number of pixel points to be 250.
  • the low-pass filter denoising process and the fuzzy channel denoising process are performed on the video image processed by the random injection retinex to obtain a brightness variation calculation formula, which specifically comprises: calculating a formula according to the relative brightness change of the pixel point based on the RSR algorithm, The luminance variation values of the three channels are calculated as the local luminance estimation vector of the RSR algorithm; and the low-pass filter and the fuzzy channel are introduced successively to improve the local luminance estimation vector of the RSR algorithm, and the luminance variation calculation after denoising processing is obtained. formula.
  • the calculating the brightness change value of the three channels by using the brightness calculation formula comprises: calculating the brightness change of any specified pixel point in each channel by using the brightness calculation formula, and calculating by using the nearest neighbor interpolation method. The brightness of the remaining pixels changes to obtain the brightness change values of the three channels.
  • the technical solution of the present invention also proposes a video image denoising and enhancement apparatus based on random injection retinex, comprising: a tuning processing module, configured to construct a set of jet pixel points, and process the random injection retinex based on the set of injected pixel points
  • the parameter involved is subjected to a tuning process, wherein the parameter includes a number of injection sets and a number of pixel points;
  • a random injection retinex processing module is configured to perform random injection retinex processing on the video image based on the parameter after the tuning process; denoising a processing module, configured to perform low-pass filter denoising processing and fuzzy channel denoising processing on the video image processed by the random injection retinex, to obtain a brightness change calculation formula; and a brightness change calculation module, configured to combine the brightness change calculation formula, Obtaining a brightness calculation formula of the output image, and calculating a brightness change value of the three channels by using the brightness calculation formula to form a local brightness estimation vector; and a channel
  • the denoising processing module comprises: an RSR algorithm processing module, configured to calculate a brightness change value of three channels according to a relative brightness change calculation formula of the pixel point based on the RSR algorithm, as a local brightness estimation vector of the RSR algorithm
  • the RSR algorithm improvement module is used to introduce a low-pass filter and a fuzzy channel successively, and improve the local luminance estimation vector of the RSR algorithm, and obtain a luminance change calculation formula after denoising processing.
  • the brightness change calculation module calculates the brightness change value of the three channels by using the brightness calculation formula, and specifically includes: calculating the brightness change of any specified pixel point in each channel by using the brightness calculation formula, and then using The nearest neighbor interpolation method calculates the brightness change of the remaining pixels, and obtains the brightness change values of the three channels.
  • the present invention proposes a brightness variation calculation formula that can effectively denoise, which greatly improves the sharpness of an image. Since the probability of the pixel points at the same angle is the same, the halo problem caused by the specific direction in the conventional algorithm is well solved.
  • the brightness calculation of the output image is used to calculate the brightness change of any pixel i in each channel, and the other pixels are calculated by the nearest neighbor interpolation method, and the calculation amount is further reduced, thereby greatly accelerating the enhancement speed, so that Enhance more frame images per unit time to meet the real-time requirements of video enhancement.
  • the invention can be widely applied to the fields of image enhancement, video analysis, target detection, tracking and recognition, and the like.
  • FIG. 1 is a flow chart showing a video image denoising and enhancement method based on random injection retinex in an embodiment of the present invention
  • 2(a) and 2(b) are analysis diagrams of ⁇ E N (n) and ⁇ E n (N) corresponding to two N values and different n values;
  • FIG. 3 is a schematic flow chart of brightness calculation in an embodiment of the present invention.
  • FIG. 4 is a schematic diagram showing the structure of a video image denoising and enhancing apparatus based on random ejection retinex in an embodiment of the present invention.
  • An embodiment of the present invention provides a video image denoising and enhancement method based on random injection retinex. As shown in FIG. 1 , the method specifically includes the following steps: constructing a set of ejection pixel points, and randomly injecting a retinex based on the set of ejection pixel points.
  • the parameters include the number of injection sets and the number of pixels; the random injection retinex processing of the video image based on the parameters after the tuning process; and the video image processed by the random injection retinex Performing low-pass filter denoising processing and fuzzy channel denoising processing to obtain a brightness change calculation formula; combining the brightness change calculation formula to obtain a brightness calculation formula of the output image, and using the brightness calculation formula to calculate the brightness change of the three channels
  • the value constitutes a local luminance estimation vector; and based on the obtained local luminance estimation vector, three-channel fusion is performed to obtain a denoised and enhanced video image.
  • the method of constructing the jet pixel points by using the polar coordinate method comprises: using a snapshot k (i) to represent n pixel points centered on the pixel point i, and the coordinates of the pixel point i are (i x , i y ), which can be set
  • the coordinates of the pixel point j belonging to the spray set Spray k (i) are (j x , j y ), and the polar coordinates are:
  • ⁇ is an arbitrary value of the interval [0, R]
  • R represents the ejection radius
  • ⁇ is an arbitrary value of the interval [0, 2 ⁇ ].
  • the probability is analyzed by the probability of the radius.
  • the probability that the pixel is in the range of radius r is r/R
  • the number of pixels is nr/R, according to the area formula According to the number formula, the A is obtained:
  • ⁇ belongs to the interval [0, R], and ⁇ is an arbitrary value of the interval [0, 2 ⁇ ].
  • the parameters involved in the random injection retinex process are tuned, wherein the parameters include a jet radius, a radius density function, a number of jet sets, and a number of pixel points.
  • the tuning of the radius density function has been explained above, and the following three main parameters are mainly described.
  • the ejection radius R defines the range of the analysis area around the pixel point i, and the obtained pixel point must be able to reflect the color information around the pixel point i.
  • the diagonal length of the video image is preferentially selected as the R value. If R is less than the diagonal length, there will definitely be two extreme pixel points in the image that will never be compared; if R is too large, there will be many pixels randomly generated in the jet set that are not in the image area.
  • the number N of jet sets and the number n of pixel points are two parameters that are closely related.
  • the analysis improves the N value of the original random injection retinex algorithm.
  • the final pixel luminance value is a weighted average of the N injection sets, and each injection set depends on how many pixels are located to find the luminance maximum.
  • n determines the amount of information obtained in the ejection area. n is too large to reflect the regionality of pixel i, too small to get enough information to reflect the surrounding color information of pixel i.
  • Fig. 2(a) and Fig. 2(b) are analysis diagrams of two N values and different n values ⁇ E N (n) and ⁇ E n (N), ⁇ E N (n) and ⁇ E n (N) A calculation function that senses differences for CIELab.
  • the N values of 1 and 5 are chosen, and n takes a continuous value of 50 between 50 and 400 steps.
  • Figure 3 is a flow chart of the source of the new brightness calculation formula. The specific steps are as follows: The calculation formula of the relative brightness change of the pixel point i:
  • C c (i) is the change in brightness of pixel i of channel c
  • I c (i) is the brightness of the original image
  • R c (i) is the brightness of the image processed by RSR.
  • the color image has three color channels, and the luminance change values of the three channels are calculated as the local luminance estimation vector [C r (i), C g (i), C b (i)] T of the RSR algorithm.
  • the effect of noise processing is not large, the difference in brightness between spatial neighboring pixels becomes smaller. It is possible to introduce a low-pass filter without interfering with the change of brightness.
  • k is the average kernel and * represents the convolution operation.
  • Another method of denoising is to introduce a fuzzy channel:
  • the brightness calculation formula O c (i) of the output image is obtained by combining the brightness change formula subjected to the denoising process:
  • k 1 and k 2 are the average nuclei.
  • the experimental average kernel has better image processing effect and execution efficiency than the Gaussian kernel.
  • k 1 k 2 .
  • the larger the average kernel size the more the noise can be removed, but at the same time the locality of the random injection retinex is reduced. On the contrary, the smaller the size, the smaller the ability to remove noise.
  • a good balancing approach is to choose a smaller core size for images that do not require significant quality enhancement, so a better choice is 25 x 25, which not only achieves the same enhanced quality, but also removes some visual anomalies seen by the human eye.
  • the pixels of the rth row and the cth column are selected (ie, any specified pixel points), and the other pixels are calculated by the nearest neighbor interpolation method.
  • the new local luminance estimation vector is [O r (i), O g (i), O b (i)] T , and the three channels are fused to obtain the final denoising and enhancement. Output video image.
  • the embodiment further provides a video image denoising and enhancement device based on random injection retinex, as shown in FIG. 4, comprising: a tuning processing module, configured to construct a jet pixel set, and based on the jet The pixel set performs a tuning process on the parameters involved in the random injection retinex processing, wherein the parameter includes the number of ejection sets and the number of pixel points; and the random injection retinex processing module is used for the video image based on the parameter after the tuning process
  • the random injection retinex processing is performed;
  • the denoising processing module is configured to perform low-pass filter denoising processing and fuzzy channel denoising processing on the video image processed by the random injection retinex to obtain a brightness change calculation formula; and a brightness change calculation module for Combined with the stated
  • the brightness change calculation formula obtains the brightness calculation formula of the output image, and uses the brightness calculation formula to calculate the brightness change values of the three channels to form a local brightness estimation vector; and the channel fusion module is configured to perform the local
  • the denoising processing module includes: an RSR algorithm processing module, configured to calculate a brightness change value of three channels according to a relative brightness change calculation formula of the pixel point based on the RSR algorithm, as a local brightness estimation vector of the RSR algorithm; And the RSR algorithm improvement module is used to introduce a low-pass filter and a fuzzy channel successively, and improve the local brightness estimation vector of the RSR algorithm, and obtain a brightness variation calculation formula after denoising processing.
  • the specific implementation process of the device corresponds to the above-mentioned random image reinex-based video image denoising and enhancement method, and will not be described here.

Landscapes

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

Abstract

本发明涉及图像处理技术领域，公开了一种基于随机喷射retinex的视频图像去噪与增强方法，包括：构建喷射像素点集合，并对随机喷射retinex处理涉及的参数进行调优处理，所述参数包含喷射集合个数和像素点个数；基于调优处理后的参数对视频图像进行随机喷射retinex处理；对随机喷射retinex处理后的视频图像进行低通滤波器去噪处理和模糊通道去噪处理，得到亮度变化计算公式；结合所述亮度变化计算公式，得到输出图像的亮度计算公式，并采用该亮度计算公式计算三个通道的亮度变化值，构成局部亮度估计矢量；以及基于得到的局部亮度估计矢量，进行三通道融合，得到去噪及增强后的视频图像。本发明提出能有效去噪的亮度变化计算公式，大大提高了图像的清晰度。

Description

基于随机喷射retinex的视频图像去噪与增强方法及装置 技术领域
本发明图像处理技术领域，特别是涉及一种基于随机喷射retinex的视频图像去噪与增强方法及装置。
背景技术
近年来，多媒体技术发展越来越成熟，随处都可见视频设备的应用，如监控、军事等，并且人们对视频画质的要求日益提高，但是在一些不可避免的客观因素作用下，视频图像的品质低下，并不能满足人们的需求。因此研究如何对视频图像进行实时的增强与去噪具有重大的意义。
视频图像品质低下主要受两方面影响:(1)一般的视频采集系统在晴天时都能捕获质量较好的图像，但是在出现大雾、沙尘等恶劣天气时，或是夜间光照不足的情况下，系统捕获的图像对比度往往很低，人们很难从画面中获取有价值的信息。利用视频图像增强技术就可以有效地改善视频图像的视觉效果，突出人们感兴趣的信息，而摒弃一些无用信息。(2)视频图像在采集或是传输过程中通常会由于噪声的污染，使得画面质量下降，视频图像偏离了真实情况，极大影响了人们从中提取信息的精确度，因此非常有必要在利用视频图像前消除噪声，提高消噪后图像的质量，突出视频图像特征。
灰色图像的增强方法较为成熟，有对比度增强方法、直方图均衡 方法、同态滤波方法、小波变换方法等。对彩色图像增强方法主要有基于Retinex理论的方法，此方法在动态范围压缩和颜色恒常性方面都具有良好的特性，如基于路径比较的retinex算法，但是这种算法存在着计算量大、对几何路径依赖性高和存在样本噪声的问题，对有噪声图像并未能取得良好效果。因此，需要改进现有技术基于Retinex理论的算法。
发明内容
本发明的目的是提供一种基于随机喷射retinex的视频图像去噪与增强方法及装置，用于解决现有技术中基于特定方向带来的光晕问题以及去噪效果不理想的问题。
为了实现上述目的，本发明的技术方案提供一种基于随机喷射retinex的视频图像去噪与增强方法，包括：构建喷射像素点集合，并基于该喷射像素点集合对随机喷射retinex处理涉及的参数进行调优处理，其中，所述参数包含喷射集合个数和像素点个数；基于调优处理后的参数对视频图像进行随机喷射retinex处理；对随机喷射retinex处理后的视频图像进行低通滤波器去噪处理和模糊通道去噪处理，得到亮度变化计算公式；结合所述亮度变化计算公式，得到输出图像的亮度计算公式，并采用该亮度计算公式计算三个通道的亮度变化值，构成局部亮度估计矢量；以及基于得到的局部亮度估计矢量，进行三通道融合，得到去噪及增强后的视频图像。
优选地，采用极坐标方式构建喷射像素点集合，具体包括：
用Sprayk(i)表示以像素点i为中心的n个像素点，像素点i的坐标 为(ix,iy)，可设属于喷射集合Sprayk(i)的像素点j的坐标为(jx,jy)，则极坐标为：
Figure PCTCN2015094088-appb-000001
其中ρ是区间[0，R]的任意值，R表示喷射半径，θ是[0,2π]区间的任意值。
优选地，对随机喷射retinex处理涉及的参数进行调优处理，其中所述参数还包括喷射半径和半径密度函数。
优选地，对随机喷射retinex处理涉及的喷射半径进行调优处理包括：选择视频图像的对角线长度为喷射半径。
优选地，对随机喷射retinex处理涉及的喷射集合个数和像素点个数进行调优处理包括：选择喷射集合个数为1，选择像素点个数为250。
优选地，对随机喷射retinex处理后的视频图像进行低通滤波器去噪处理和模糊通道去噪处理，得到亮度变化计算公式，具体包括：基于RSR算法，根据像素点的相对亮度变化计算公式，计算出3个通道的亮度变化值，作为RSR算法的局部亮度估计矢量；以及先后引入低通滤波器和模糊通道，对RSR算法的局部亮度估计矢量进行改进，得到经过去噪处理的亮度变化计算公式。
优选地，所述采用该亮度计算公式计算三个通道的亮度变化值，具体包括：采用该亮度计算公式计算每个通道中的任意指定的像素点的亮度变化，再使用最近邻域插值法计算其余像素点的亮度变化，得到三个通道的亮度变化值。
本发明的技术方案还提出了一种基于随机喷射retinex的视频图像去噪与增强装置，包括：调优处理模块，用于构建喷射像素点集合，并基于该喷射像素点集合对随机喷射retinex处理涉及的参数进行调优处理，其中，所述参数包含喷射集合个数和像素点个数；随机喷射retinex处理模块，用于基于调优处理后的参数对视频图像进行随机喷射retinex处理；去噪处理模块，用于对随机喷射retinex处理后的视频图像进行低通滤波器去噪处理和模糊通道去噪处理，得到亮度变化计算公式；亮度变化计算模块，用于结合所述亮度变化计算公式，得到输出图像的亮度计算公式，并采用该亮度计算公式计算三个通道的亮度变化值，构成局部亮度估计矢量；以及通道融合模块，用于基于得到的局部亮度估计矢量，进行三通道融合，得到去噪及增强后的视频图像。
优选地，所述去噪处理模块包括：RSR算法处理模块，用于基于RSR算法，根据像素点的相对亮度变化计算公式，计算出3个通道的亮度变化值，作为RSR算法的局部亮度估计矢量；以及RSR算法改进模块，用于先后引入低通滤波器和模糊通道，对RSR算法的局部亮度估计矢量进行改进，得到经过去噪处理的亮度变化计算公式。
优选地，所述亮度变化计算模块中，采用该亮度计算公式计算三个通道的亮度变化值，具体包括：采用该亮度计算公式计算每个通道中的任意指定的像素点的亮度变化，再使用最近邻域插值法计算其余像素点的亮度变化，得到三个通道的亮度变化值。
本发明的有益效果是：
1、本发明提出能有效去噪的亮度变化计算公式，大大提高了图像的清晰度。由于像素点在各角度上的概率相同，因此很好的解决了传统算法中由于特定方向带来的光晕问题。
2、本发明降低了算法复杂度，优化喷射集个数和像素点个数，取N＝1和n＝250，运行速度得到进一步提高。
3、本发明对输出图像的亮度计算采用计算每个通道中任意像素点i的亮度变化，其他像素点用最近邻域插值法计算方法，运算量进一步减少，大大加快增强速度，这样就能在单位时间内增强更多帧图像，满足视频增强的实时性要求。
4、本发明可广泛应用于图像增强、视频分析、目标检测、跟踪和识别的预处理等领域。
本发明的其它特征和优点将在随后的具体实施方式部分予以详细说明。
附图说明
附图是用来提供对本发明的进一步理解，并且构成说明书的一部分，与下面的具体实施方式一起用于解释本发明，但并不构成对本发明的限制。在附图中：
图1是本发明的实施方式中基于随机喷射retinex的视频图像去噪与增强方法的流程示意图；
图2(a)和图2(b)是两种N值和不同n值对应的△EN(n)和△En(N) 的分析图；
图3是本发明的实施方式中亮度计算的流程示意图；
图4是本发明的实施方式中基于随机喷射retinex的视频图像去噪与增强装置的结构示意图。
具体实施方式
以下结合附图对本发明的具体实施方式进行详细说明。应当理解的是，此处所描述的具体实施方式仅用于说明和解释本发明，并不用于限制本发明。
本发明的实施方式提出了一种基于随机喷射retinex的视频图像去噪与增强方法，如图1所示，具体包括以下步骤：构建喷射像素点集合，并基于该喷射像素点集合对随机喷射retinex处理涉及的参数进行调优处理，其中，所述参数包含喷射集合个数和像素点个数；基于调优处理后的参数对视频图像进行随机喷射retinex处理；对随机喷射retinex处理后的视频图像进行低通滤波器去噪处理和模糊通道去噪处理，得到亮度变化计算公式；结合所述亮度变化计算公式，得到输出图像的亮度计算公式，并采用该亮度计算公式计算三个通道的亮度变化值，构成局部亮度估计矢量；以及基于得到的局部亮度估计矢量，进行三通道融合，得到去噪及增强后的视频图像。
基于图1所示的各个基本步骤，本实施方式的具体实施过程分为以下几个部分。
一、构建喷射像素点集合，进行参数调优，并基于调优后的参数 对视频图像进行随机喷射retinex处理。
其中，采用极坐标方式构建喷射像素点集合，具体包括：用Sprayk(i)表示以像素点i为中心的n个像素点，像素点i的坐标为(ix,iy)，可设属于喷射集合Sprayk(i)的像素点j的坐标为(jx,jy)，则极坐标为：
Figure PCTCN2015094088-appb-000002
其中ρ是区间[0，R]的任意值，R表示喷射半径，θ是[0,2π]区间的任意值。
从上式中可以分析得到像素点落在各个角度上是方向概率是相同的，因此以半径大小的概率进行分析。对于半径为r的区域Cr，面积为A＝πr2，像素点在半径r范围内的概率是r/R，因此像素点个数为nr/R，根据面积公式个数为
Figure PCTCN2015094088-appb-000003
根据个数公式对A求导得：
Figure PCTCN2015094088-appb-000004
带入A＝πr2，得
Figure PCTCN2015094088-appb-000005
由此公式可以认为随半径增大，区域像素点个数是减小的。通过对极坐标的ρ构建函数f，可以把极坐标式变换为：
Figure PCTCN2015094088-appb-000006
其中ρ属于区间[0，R]，θ是[0,2π]区间的任意值。对于log函数和双曲正弦函数分布效果十分近似于f(ρ)＝ρ的密度函数分布；指数函数中，指数大于1时指数值增大中心区域密度增大，指数小于1时使 指数值减小越加偏离原始区域；线性函数是对区域半径的一个比例扩大或者缩小的作用。从计算量和作用效果来说取f(ρ)＝ρ为最佳半径密度函数，实现了对半径密度函数的调优。
另外，对随机喷射retinex处理涉及的参数进行调优处理，其中所述参数包括喷射半径、半径密度函数、喷射集合个数和像素点个数。关于半径密度函数的调优已在上文中说明，下面主要介绍另外三个参数。
喷射半径R：定义的是像素点i周边的分析区域范围，得到的像素点必须要能够反应像素点i周边的颜色信息，本实施方式中优先选择视频图像的对角线长度为R值。若R小于对角线长度，则图像中肯定会有两个极端像素点永远不会被进行比较运算；若R取得太大，则喷射集合随机产生的像素点会有很多不在图像区域上。
喷射集合个数N和像素点个数n是密切相关的两个参数。分析改进原随机喷射retinex算法的N取值。最终的像素亮度值是对N个喷射集合的加权平均，而每个喷射集取决于有多少个像素点来寻找亮度最大值。n决定的是喷射区域中得到的信息量大小。n太大不能体现像素点i的区域性，太小又不能得到足够信息量来反映像素点i的周边颜色信息。
最佳N和n值是通过降低N·n值获得。图2(a)和图2(b)是两种N值和不同n值△EN(n)和△En(N)的分析图，△EN(n)和△En(N)为CIELab感知差异的计算函数。选择1和5的N值，n取50到400之间步阶为50的连续值。在约束条件△En(N)<1和△EN(n)<1下，由图2 可得1×250＝250<5×250＝1250，所以满足约束条件的取N＝1和n＝250，比现有技术中随机喷射retinex提出的N＝20，n＝400值在去除噪声后稳定性更好，也减小了算法的复杂度。
二、去噪处理
图3是新的亮度计算公式来源流程图，具体步骤如下：像素点i的相对亮度变化计算公式：
Figure PCTCN2015094088-appb-000007
其中Cc(i)是通道c的像素点i的亮度变化，Ic(i)是原图像亮度，Rc(i)是用RSR处理的图像亮度。彩色图像有3个颜色通道，则计算出3个通道的亮度变化值，作为RSR算法的局部亮度估计矢量[Cr(i),Cg(i),Cb(i)]T。尽管对噪声的处理效果不大，但空间邻域像素之间的亮度差异变小了，在不干预亮度变化条件下，引入低通滤波器变成了可能，则有下式：
Cc,k(i)＝(Cc*k)(i)
其中k是平均核，*代表卷积运算。
另一种去噪方法是引入模糊通道：
Figure PCTCN2015094088-appb-000008
为得到更好的去噪效果，将引入低通滤波器和引入模糊通道结合在一起，得到新的亮度变化计算公式：
Figure PCTCN2015094088-appb-000009
三、亮度变化计算及通道融合
结合经过去噪处理的亮度变化公式得到输出图像的亮度计算公式Oc(i)为：
Figure PCTCN2015094088-appb-000010
其中k1和k2是平均核。经试验平均核比高斯核有更好的图像处理效果和执行效率。为简化参数调优，令k1＝k2。平均核尺寸越大越能去除噪声，但同时减少了随机喷射retinex的局部性，相反尺寸越小，去除噪声的能力越小。一个好的平衡办法是对无需重要质量增强的图像选择较小核尺寸，所以较好的选择是25×25，不仅能获得相同的增强质量，并且能去除人眼看到的一些视觉异常现象。
为提高视频图像处理的实时性，这里不是对每个像素计算亮度变化，而选择处理第r行和第c列的像素(即任意指定的像素点)，其他像素则用最近邻域插值法计算，优选为r＝c＝1。
最后，计算3个通道的亮度变化值，则新的局部亮度估计矢量为[Or(i),Og(i),Ob(i)]T，将3通道融合得到最终去噪和增强的输出视频图像。
对应地，本实施方式还给出了一种基于随机喷射retinex的视频图像去噪与增强装置，如图4所示，包括：调优处理模块，用于构建喷射像素点集合，并基于该喷射像素点集合对随机喷射retinex处理涉及的参数进行调优处理，其中，所述参数包含喷射集合个数和像素点个数；随机喷射retinex处理模块，用于基于调优处理后的参数对视频图像进行随机喷射retinex处理；去噪处理模块，用于对随机喷射retinex处理后的视频图像进行低通滤波器去噪处理和模糊通道去噪处理，得到亮度变化计算公式；亮度变化计算模块，用于结合所述 亮度变化计算公式，得到输出图像的亮度计算公式，并采用该亮度计算公式计算三个通道的亮度变化值，构成局部亮度估计矢量；以及通道融合模块，用于基于得到的局部亮度估计矢量，进行三通道融合，得到去噪及增强后的视频图像。
其中，所述去噪处理模块包括：RSR算法处理模块，用于基于RSR算法，根据像素点的相对亮度变化计算公式，计算出3个通道的亮度变化值，作为RSR算法的局部亮度估计矢量；以及RSR算法改进模块，用于先后引入低通滤波器和模糊通道，对RSR算法的局部亮度估计矢量进行改进，得到经过去噪处理的亮度变化计算公式。
该装置的具体实施过程与上述的基于随机喷射retinex的视频图像去噪与增强方法一一对应，这里不再多述。
以上结合附图详细描述了本发明的优选实施方式，但是，本发明并不限于上述实施方式中的具体细节，在本发明的技术构思范围内，可以对本发明的技术方案进行多种简单变型，这些简单变型均属于本发明的保护范围。
另外需要说明的是，在上述具体实施方式中所描述的各个具体技术特征，在不矛盾的情况下，可以通过任何合适的方式进行组合，为了避免不必要的重复，本发明对各种可能的组合方式不再另行说明。
此外，本发明的各种不同的实施方式之间也可以进行任意组合，只要其不违背本发明的思想，其同样应当视为本发明所公开的内容。

Claims (10)

  1. 一种基于随机喷射retinex的视频图像去噪与增强方法，其特征在于，包括：
    构建喷射像素点集合，并基于该喷射像素点集合对随机喷射retinex处理涉及的参数进行调优处理，其中，所述参数包含喷射集合个数和像素点个数；
    基于调优处理后的参数对视频图像进行随机喷射retinex处理；
    对随机喷射retinex处理后的视频图像进行低通滤波器去噪处理和模糊通道去噪处理，得到亮度变化计算公式；
    结合所述亮度变化计算公式，得到输出图像的亮度计算公式，并采用该亮度计算公式计算三个通道的亮度变化值，构成局部亮度估计矢量；以及
    基于得到的局部亮度估计矢量，进行三通道融合，得到去噪及增强后的视频图像。
  2. 根据权利要求1所述的视频图像去噪与增强方法，其特征在于，采用极坐标方式构建喷射像素点集合，具体包括：
    用Sprayk(i)表示以像素点i为中心的n个像素点，像素点i的坐标为(ix,iy)，可设属于喷射集合Sprayk(i)的像素点j的坐标为(jx,jy)，则极坐标为：
    Figure PCTCN2015094088-appb-100001
    其中ρ是区间[0，R]的任意值，R表示喷射半径，θ是[0,2π]区间的 任意值。
  3. 根据权利要求1所述的视频图像去噪与增强方法，其特征在于，对随机喷射retinex处理涉及的参数进行调优处理，其中所述参数还包括喷射半径和半径密度函数。
  4. 根据权利要求3所述的视频图像去噪与增强方法，其特征在于，对随机喷射retinex处理涉及的喷射半径进行调优处理包括：选择视频图像的对角线长度为喷射半径。
  5. 根据权利要求1所述的视频图像去噪与增强方法，其特征在于，对随机喷射retinex处理涉及的喷射集合个数和像素点个数进行调优处理包括：选择喷射集合个数为1，选择像素点个数为250。
  6. 根据权利要求1所述的视频图像去噪与增强方法，其特征在于，对随机喷射retinex处理后的视频图像进行低通滤波器去噪处理和模糊通道去噪处理，得到亮度变化计算公式，具体包括：
    基于RSR算法，根据像素点的相对亮度变化计算公式，计算出3个通道的亮度变化值，作为RSR算法的局部亮度估计矢量；以及
    先后引入低通滤波器和模糊通道，对RSR算法的局部亮度估计矢量进行改进，得到经过去噪处理的亮度变化计算公式。
  7. 根据权利要求1所述的视频图像去噪与增强方法，其特征在于，所述采用该亮度计算公式计算三个通道的亮度变化值，具体包括：采用该亮度计算公式计算每个通道中的任意指定的像素点的亮度变化，再使用最近邻域插值法计算其余像素点的亮度变化，得到三个通道的亮度变化值。
  8. 一种基于随机喷射retinex的视频图像去噪与增强装置，其特征在于，包括：
    调优处理模块，用于构建喷射像素点集合，并基于该喷射像素点集合对随机喷射retinex处理涉及的参数进行调优处理，其中，所述参数包含喷射集合个数和像素点个数；
    随机喷射retinex处理模块，用于基于调优处理后的参数对视频图像进行随机喷射retinex处理；
    去噪处理模块，用于对随机喷射retinex处理后的视频图像进行低通滤波器去噪处理和模糊通道去噪处理，得到亮度变化计算公式；
    亮度变化计算模块，用于结合所述亮度变化计算公式，得到输出图像的亮度计算公式，并采用该亮度计算公式计算三个通道的亮度变化值，构成局部亮度估计矢量；以及
    通道融合模块，用于基于得到的局部亮度估计矢量，进行三通道融合，得到去噪及增强后的视频图像。
  9. 根据权利要求8所述的视频图像去噪与增强装置，其特征在于，所述去噪处理模块包括：
    RSR算法处理模块，用于基于RSR算法，根据像素点的相对亮度变化计算公式，计算出3个通道的亮度变化值，作为RSR算法的局部亮度估计矢量；以及
    RSR算法改进模块，用于先后引入低通滤波器和模糊通道，对RSR算法的局部亮度估计矢量进行改进，得到经过去噪处理的亮度变化计算公式。
  10. 根据权利要求8所述的视频图像去噪与增强装置，其特征在于，所述亮度变化计算模块中，采用该亮度计算公式计算三个通道的亮度变化值，具体包括：采用该亮度计算公式计算每个通道中的任意指定的像素点的亮度变化，再使用最近邻域插值法计算其余像素点的亮度变化，得到三个通道的亮度变化值。
PCT/CN2015/094088 2015-04-20 2015-11-09 基于随机喷射retinex的视频图像去噪与增强方法及装置 Ceased WO2016169244A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US15/305,350 US9978131B2 (en) 2015-04-20 2015-11-09 Video image denoising and enhancing method and device based on random spray retinex

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201510187842.7A CN104796582B (zh) 2015-04-20 2015-04-20 基于随机喷射retinex的视频图像去噪与增强方法及装置
CN201510187842.7 2015-04-20

Publications (1)

Publication Number Publication Date
WO2016169244A1 true WO2016169244A1 (zh) 2016-10-27

Family

ID=53561080

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2015/094088 Ceased WO2016169244A1 (zh) 2015-04-20 2015-11-09 基于随机喷射retinex的视频图像去噪与增强方法及装置

Country Status (3)

Country Link
US (1) US9978131B2 (zh)
CN (1) CN104796582B (zh)
WO (1) WO2016169244A1 (zh)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111260589A (zh) * 2020-02-20 2020-06-09 国网陕西省电力公司电力科学研究院 基于Retinex的输电线路监控图像去雾方法
CN111275643A (zh) * 2020-01-20 2020-06-12 西南科技大学 基于通道和空间注意力的真实噪声盲去噪网络模型及方法
CN111784601A (zh) * 2020-06-27 2020-10-16 浙江同善人工智能技术有限公司 一种图像去雾方法

Families Citing this family (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104796582B (zh) 2015-04-20 2018-05-04 中国矿业大学 基于随机喷射retinex的视频图像去噪与增强方法及装置
CN111046887A (zh) * 2018-10-15 2020-04-21 华北电力大学（保定） 一种带噪图像特征提取的方法
CN109472755A (zh) * 2018-11-06 2019-03-15 武汉高德智感科技有限公司 一种红外图像对数log域增强方法
CN109747824B (zh) * 2019-01-14 2023-08-25 中国计量大学 用于烟囱内部无人机避障的装置和避障方法
CN110675343B (zh) * 2019-09-24 2023-02-24 西安科技大学 一种井下输煤图像的图像增强方法
CN110706174B (zh) * 2019-09-27 2022-03-18 集美大学 一种图像增强方法、终端设备及存储介质
US12475377B2 (en) * 2021-02-05 2025-11-18 Microsoft Technology Licensing, Llc Neural command line interface example generation
CN113284200A (zh) * 2021-05-19 2021-08-20 电子科技大学成都学院 一种基于5g通信及图像识别的标识编码生成系统及方法
CN115393236B (zh) * 2022-07-12 2025-08-01 中南大学 一种幽光高温工业内窥镜取像方法及系统
CN115861106A (zh) * 2022-12-07 2023-03-28 广东方柚科技有限公司 一种夜间微光成像方法、系统、电子设备及介质
CN119048384B (zh) * 2024-08-05 2025-11-04 四川国创新视超高清视频科技有限公司 基于去噪的4k暗光视频增强方法及系统

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2005182232A (ja) * 2003-12-17 2005-07-07 Matsushita Electric Ind Co Ltd 輝度補正装置および輝度補正方法
CN101303766A (zh) * 2008-07-09 2008-11-12 北京航空航天大学 一种基于Retinex理论的快速彩色图像增强方法
KR20100027888A (ko) * 2008-09-03 2010-03-11 현대자동차주식회사 레티넥스 기법을 이용한 영상처리방법
CN102682436A (zh) * 2012-05-14 2012-09-19 陈军 一种改进的多尺度Retinex理论的图像增强方法
CN103295205A (zh) * 2013-06-25 2013-09-11 安科智慧城市技术（中国）有限公司 一种基于Retinex的微光图像快速增强方法和装置
CN104346776A (zh) * 2013-08-02 2015-02-11 杭州海康威视数字技术股份有限公司 基于Retinex理论的非线性图像增强方法及其系统
CN104796582A (zh) * 2015-04-20 2015-07-22 中国矿业大学 基于随机喷射retinex的视频图像去噪与增强方法及装置

Family Cites Families (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101303756A (zh) 2007-05-11 2008-11-12 北京五洲阳光国际科技有限公司 因特网大型多人在线三维电子商务系统和方法
CN101615244A (zh) * 2008-06-26 2009-12-30 上海梅山钢铁股份有限公司 手写板坯号自动识别方法及识别装置
CN102663697B (zh) * 2012-04-01 2014-04-30 大连海事大学 一种水下彩色视频图像的增强方法
CN103387123A (zh) * 2012-05-09 2013-11-13 刘晓奇 基于机器视觉的矿用胶带撕裂智能检测系统
CN102999885A (zh) * 2012-09-25 2013-03-27 广东威创视讯科技股份有限公司 Retinex视频增强算法亮度平均值确定方法与装置
CN103268598B (zh) * 2013-06-13 2015-09-16 武汉大学 基于Retinex理论的低照度低空遥感影像增强方法
CN104050642B (zh) * 2014-06-18 2017-01-18 上海理工大学 彩色图像复原方法
CN104091310A (zh) * 2014-06-24 2014-10-08 三星电子（中国）研发中心 一种图像去雾方法和装置

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2005182232A (ja) * 2003-12-17 2005-07-07 Matsushita Electric Ind Co Ltd 輝度補正装置および輝度補正方法
CN101303766A (zh) * 2008-07-09 2008-11-12 北京航空航天大学 一种基于Retinex理论的快速彩色图像增强方法
KR20100027888A (ko) * 2008-09-03 2010-03-11 현대자동차주식회사 레티넥스 기법을 이용한 영상처리방법
CN102682436A (zh) * 2012-05-14 2012-09-19 陈军 一种改进的多尺度Retinex理论的图像增强方法
CN103295205A (zh) * 2013-06-25 2013-09-11 安科智慧城市技术（中国）有限公司 一种基于Retinex的微光图像快速增强方法和装置
CN104346776A (zh) * 2013-08-02 2015-02-11 杭州海康威视数字技术股份有限公司 基于Retinex理论的非线性图像增强方法及其系统
CN104796582A (zh) * 2015-04-20 2015-07-22 中国矿业大学 基于随机喷射retinex的视频图像去噪与增强方法及装置

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111275643A (zh) * 2020-01-20 2020-06-12 西南科技大学 基于通道和空间注意力的真实噪声盲去噪网络模型及方法
CN111260589A (zh) * 2020-02-20 2020-06-09 国网陕西省电力公司电力科学研究院 基于Retinex的输电线路监控图像去雾方法
CN111784601A (zh) * 2020-06-27 2020-10-16 浙江同善人工智能技术有限公司 一种图像去雾方法
CN111784601B (zh) * 2020-06-27 2024-05-14 浙江同善人工智能技术有限公司 一种图像去雾方法

Also Published As

Publication number Publication date
US9978131B2 (en) 2018-05-22
US20180053287A1 (en) 2018-02-22
CN104796582B (zh) 2018-05-04
CN104796582A (zh) 2015-07-22

Similar Documents

Publication Publication Date Title
CN104796582B (zh) 基于随机喷射retinex的视频图像去噪与增强方法及装置
CN109360156B (zh) 基于生成对抗网络的图像分块的单张图像去雨方法
Varghese et al. Video denoising based on a spatiotemporal Gaussian scale mixture model
Rajkumar et al. A comparative analysis on image quality assessment for real time satellite images
CN105096280B (zh) 处理图像噪声的方法及装置
CN103873743A (zh) 一种基于结构张量和卡尔曼滤波的视频去噪方法
CN105006001B (zh) 一种基于非线性结构相似度偏差的有参图像的质量评估方法
CN109447903B (zh) 一种半参考型超分辨率重构图像质量评价模型的建立方法
Tao et al. LEGAN: A low-light image enhancement generative adversarial network for industrial internet of smart-cameras
CN111598791B (zh) 一种基于改进动态大气散射系数函数的图像去雾方法
CN106991670B (zh) 一种无参考噪声图像质量评价方法及系统
CN115249211B (zh) 一种基于水下非均匀入射光模型的图像复原方法
Yu et al. Image and video dehazing using view-based cluster segmentation
CN106530277B (zh) 一种基于小波方向相关系数的图像融合方法
Yin et al. Infrared and visible image fusion based on NSCT and fuzzy logic
CN106651789B (zh) 一种面向压缩人脸图像的自适应去块效应方法
CN109003247B (zh) 彩色图像混合噪声的去除方法
CN113781333A (zh) 一种基于引导滤波的gan网络对水下图像进行处理的方法
CN111461999B (zh) 一种基于超像素相似性测量的sar图像相干斑抑制方法
Hua et al. Low-light image enhancement based on joint generative adversarial network and image quality assessment
Zhang et al. Dehazing with improved heterogeneous atmosphere light estimation and a nonlinear color attenuation prior model
CN113838023A (zh) 一种色差图像质量评价方法
TWI460667B (zh) 模糊指紋影像重建方法
Panguluri et al. Efficient dwt based fusion algorithm for improving contrast and edge preservation
CN104463813A (zh) 一种基于噪声识别的红外图像降噪方法

Legal Events

Date Code Title Description
WWE Wipo information: entry into national phase

Ref document number: 15305350

Country of ref document: US

121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 15889727

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: 15889727

Country of ref document: EP

Kind code of ref document: A1