Nothing Special   »   [go: up one dir, main page]

CN101087365B - A method for filtering image mixed noise - Google Patents

A method for filtering image mixed noise Download PDF

Info

Publication number
CN101087365B
CN101087365B CN2006100919045A CN200610091904A CN101087365B CN 101087365 B CN101087365 B CN 101087365B CN 2006100919045 A CN2006100919045 A CN 2006100919045A CN 200610091904 A CN200610091904 A CN 200610091904A CN 101087365 B CN101087365 B CN 101087365B
Authority
CN
China
Prior art keywords
pixel
noise
image
filtering
impulsive noise
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.)
Expired - Fee Related
Application number
CN2006100919045A
Other languages
Chinese (zh)
Other versions
CN101087365A (en
Inventor
曹刚
徐立峰
罗宏宇
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.)
Cheng Guihua
Original Assignee
ZTE Corp
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 ZTE Corp filed Critical ZTE Corp
Priority to CN2006100919045A priority Critical patent/CN101087365B/en
Publication of CN101087365A publication Critical patent/CN101087365A/en
Application granted granted Critical
Publication of CN101087365B publication Critical patent/CN101087365B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Landscapes

  • Image Processing (AREA)
  • Facsimile Image Signal Circuits (AREA)

Abstract

The invention discloses a method of filtering image mixed noise, it includes step 1: selecting input first pixel in original image which includes mixed noise; step 2: judging whether the pixel belongs to pulse noise, if it is, enter step 3. if else enter step 4; step 3: the pixel is median filter processed to filter the pulse noise, then enter step 6; step 4: ensuring the members of the un-pulse noise pixel; step 5: gauss noise is filtered by filter; step 6: checking whether the all pixels are processed, if it is not, the next pixel is fetched, then turn to step 2; if it is, new image whose mixed noise is filtered is output. The method can filter mixed noise in image, at the same time edge and detail information of image can be protected well, and calculation speed is high.

Description

A kind of method of filtering image mixed noise
Technical field:
The present invention relates to a kind of method of filtering image mixed noise in the image processing field, relate in particular in the realtime graphic process field filtering picture noise.
Background technology:
In image processing field, it is very important that the filtering of picture noise is handled for successive image.In existing image de-noising method commonly used, the gaussian filtering method is more effective to filtering image Gaussian noise, but damages image edge information easily simultaneously, thereby image is thickened.And the median filtering method impulsive noise of filtering image well, but for filtering image Gaussian noise poor effect then.Both common defectives are that all neighborhood territory pixels in the filter window are all adopted identical processing, thereby can introduce error, cause the damage of image border and detailed information.
Filter method is a kind of image de-noising method that can protect image border and detailed information well on the same group.The main thought of this method is only to find out neighborhood territory pixel close with the center pixel distance feature in the filter window as participating in filtering with the group membership, its process is to carry out the ascending order arrangement according to neighborhood territory pixel in the filter window and former pixel characteristic distance value earlier, differentiate the same group membership who finds out this pixel by Fisher again, replace original pixel characteristic value with the weighted feature value that belongs in the filter window with group membership's pixel then.This shows that filtering on the same group can regard a kind of gaussian filtering that has binary mask as, wherein 1 this filter window neighborhood territory pixel of expression is that the same group membership of former pixel will participate in filtering, and 0 represents it is not participate in filtering with the group membership.Filtering has overcome the drawback that all pixels in the filter window of gaussian filtering in the past all participate in filtering on the same group, thus well filtering the image Gaussian noise, can protect edge of image and detailed information again.But the filtering impulsive noise of filtering image effectively on the same group, its amount of calculation is very big when sorting out the same group membership of each pixel simultaneously, particularly when filter window was chosen greatly, its denoising speed was extremely slow, thereby this method is not suitable for the realtime graphic processing very much.And generally, original image always comprises two kinds of noises the most common, i.e. impulsive noise and Gaussian noise.So the mixed noise of filtering image apace, the image de-noising method that can protect image border and detailed information again becomes the research focus in the present image processing field.
Summary of the invention
The technical problem to be solved in the present invention is to propose a kind of method of filtering image mixed noise; this method fast and effeciently contains the mixed noise of gaussian sum pulse in the filtering image; simultaneously can protect edge of image and detailed information well again, be particularly useful for the processing of realtime graphic.
Method of filtering image mixed noise of the present invention comprises:
Step 1: the first pixel in original image that contains mixed noise of choosing input;
Step 2: differentiate this pixel and whether belong to impulsive noise,, then enter step 3, otherwise directly enter step 4 if this pixel belongs to impulsive noise;
Step 3: this pixel is carried out medium filtering handle, this impulsive noise is got rid of, directly enter step 6 then;
Step 4: the same group membership who determines this non-pulse noise pixel;
Step 5: by filtering filtering Gaussian noise on the same group;
Step 6: whether all pixels of check image dispose, and then obtain the next pixel of image if be untreated, and step 2 is returned in redirect then; If handle the new images of then exporting behind the filtering mixed noise;
Wherein, the described classification with the group membership of described step 4 further comprises: calculate this pixel grey scale second differnce in the horizontal direction; Calculate this pixel grey scale second differnce in vertical direction; Calculate the two-dimentional deviation branch of this pixel grey scale; Calculate the direction of this pixel grey scale changing features minimum; In this pixel filter window, obtain the same group membership that neighborhood territory pixel on this minimum direction constitutes this pixel.
Compare with existing method, adopting the method for the invention is to judge whether pending pixel belongs to impulsive noise earlier, if impulsive noise is then got rid of by medium filtering, otherwise carry out same group membership's classification of this pixel, carry out filtering on the same group then and filter out Gaussian noise.The inventive method is the mixed noise in the filtering image effectively, can protect edge of image and detailed information well simultaneously, and the arithmetic speed of its denoising is also higher.
Description of drawings:
Fig. 1 is the overview flow chart of the filtering image mixed noise method that proposes of the present invention.
Fig. 2 is the present invention differentiates impulsive noise by calculating pixel local energy value a flow chart.
Fig. 3 is the flow chart that the present invention sorts out the member method on the same group of pixel.
Embodiment:
Below in conjunction with accompanying drawing and embodiment method of the present invention is described, Fig. 1 is the overview flow chart of the filtering image mixed noise method that proposes of the present invention.
Step 1: the first pixel in original image that contains mixed noise of choosing input.
Step 2: differentiate this pixel and whether belong to impulsive noise.The method of differentiating usually can by calculate this pixel level, vertically or the feature difference methods such as (as gray scale, rgb values etc.) on the gradient direction differentiate impulsive noise, also can judge whether this pixel is impulsive noise according to the local energy feature of pixel.In the present embodiment, to judge according to the local energy feature of pixel whether this pixel is that impulsive noise is that example describes.The local energy feature of pixel has mainly been portrayed the local difference information of pixel characteristic value on level, a plurality of directions such as vertical, and the local energy value of impulsive noise respective pixel is often obviously than the local energy value height of image smoothing zone or fringe region pixel usually.Therefore can effectively judge impulsive noise by the local energy value of calculating pixel, and calculate simply fast, protect edge of image and detailed information again well.
If this pixel belongs to impulsive noise, then enter following step 3, otherwise directly enter step 4.
Step 3: this pixel is carried out medium filtering handle, this impulsive noise is got rid of, directly enter step 6 then.
Step 4: the same group membership who determines this non-pulse noise pixel.Before using the Gaussian noise of filtering filtering image on the same group, a step of most critical is that the neighborhood territory pixel in this pixel filter window is carried out sorting out with the group membership, promptly determines the same group membership of this pixel.Here so-calledly be meant one group of close with the center pixel feature in filter window neighborhood territory pixel with the group membership.Determine that pixel has multiple with group membership's method, relatively Chang Yong method is to calculate earlier the distance of the characteristic vector of each neighborhood territory pixel and center pixel in filter window, then these distance values are carried out ascending order and arrange, adopt Fisher to differentiate at last and come definite member on the same group.Also having a kind of method practical, the same group membership of definite this pixel quickly is directly will be classified as the same group membership of this pixel at the neighborhood territory pixel on this pixel characteristic gradient vertical direction (being the minimum direction of changing features) in filter window, and wherein the minimum direction of the changing features of pixel can be extracted exactly by the feature difference information method of calculating this pixel.The amount of calculation of this method is very little, can save the classification time with the group membership greatly.
Step 5: by filtering filtering Gaussian noise on the same group.After the same group membership of this pixel determines, can come the filtering Gaussian noise by filtering on the same group.The main processing procedure of filtering is the same group membership's of pixel weighted feature value to be substituted the original characteristic value of former this pixel on the same group, thereby reaches the purpose of removing noise.
Step 6: whether all pixels of check image dispose.Then obtain the next pixel of image if be untreated, step 2 is returned in redirect then; If handle the new images of then exporting behind the filtering mixed noise.
Below in conjunction with Fig. 2 and Fig. 3, coming filtering one size with method of the present invention is that pulse and the Gaussian Mixture noise that contains in the gray level image (with m=256) of m*m is that example further describes, and its concrete steps are as follows:
1 according to image from left to right, from top to bottom scanning sequency obtains the first pixel in original image that contains mixed noise;
Whether 2 differentiate this pixel belongs to impulsive noise.Fig. 2 is a flow chart of differentiating impulsive noise by calculating pixel local energy value, and its specific implementation process is as follows:
(coordinate is (x to 2-1 with this test point pixel, y)) be the center, choose the filter window of a n*n (n=9 herein) size, this window size choose complexity and the accuracy that is related to calculating, window is big more, and calculation times will increase but accuracy also can increase;
Pixel grey scale average μ in the 2-2 calculation of filtered window:
μ = 1 9 * 9 Σ i = - 4 4 Σ j = - 4 4 p ( x + i , y + j ) - - - ( 1 )
(x is that image coordinate is (x, y) gray value of pixel y) to p in the formula;
Formula came the local energy of calculating pixel below 2-3 adopted:
E(x,y)=|2*(p(x,y)-μ) 2-(p(x-1,y)-μ)*(p(x+1,y)-μ)
-(p(x,y-1)-μ)*(p(x,y+1)-μ)| (2)
The local energy average that 2-4 chooses all pixels of image is used as the local energy threshold value, and this threshold value Th is calculated as follows:
Th = 1 256 * 256 × Σ n = 1 256 * 256 E n - - - ( 3 )
E in the formula nThe local energy value of n pixel in the presentation video;
2-5 judges according to the local energy threshold value (whether x is impulsive noise y), even to this pixel at last
E(x,y)>Th (4)
Judge that then it is an impulsive noise, the 3rd step below carrying out.Otherwise be not impulsive noise, directly carry out next the 4th step.
3 paired pulses noise pixels carry out medium filtering to be handled, and the filter window size is still chosen 9*9, carries out for the 6th step then.
4 pairs of non-pulse noise pixels carry out sorting out with the group membership, and Fig. 3 is for sorting out the flow chart of this pixel with the group membership, and it is implemented as follows:
4-1 calculates this pixel grey scale second differnce in the horizontal direction earlier, that is:
Δ x 2 p ( x , y ) = Δ x p ( x + 1 , y ) - Δ x p ( x , y ) - - - ( 5 )
Wherein:
Δ xp(x,y)=p(x+1,y)-p(x,y) (6)
4-2 calculates this pixel grey scale second differnce in vertical direction again, that is:
Δ y 2 p ( x , y ) = Δ y p ( x , y + 1 ) - Δ y p ( x , y ) - - - ( 7 )
Wherein:
Δ yp(x,y)=p(x,y+1)-p(x,y) (8)
4-3 calculates the two-dimentional deviation branch of this pixel grey scale then, that is:
Δ xyp(x,y)=Δ xp(x,y+1)-Δ xp(x,y) (9)
4-4 calculates the direction ξ of this pixel grey scale changing features minimum again, that is:
ξ = π 2 + 1 2 arctan ( 2 Δ xy p ( x , y ) Δ x 2 p ( x , y ) - Δ y 2 p ( x , y ) ) - - - ( 10 )
4-5 obtains the same group membership that neighborhood territory pixel on this minimum direction ξ constitutes this pixel at last in this pixel filter window.
5 uses are carried out filtering on the same group with the group membership and are come the filtering Gaussian noise.After the same group membership of this pixel determines, substitute the gray feature value of former pixel with regard to available with the weighting gray feature value of group membership's pixel, that is:
p new ( x , y ) = Σ i = 0 s - 1 w i p i ( x , y ) Σ i = 0 s - 1 w i - - - ( 11 )
P wherein i(x y) belongs to the same group membership's of this detection pixel gray feature vector, w iBe the weight coefficient of corresponding gaussian filtering, s is the number with the group membership.
Whether all pixels of 6 check image dispose by above-mentioned steps, if be untreated then still according to image from left to right, from top to bottom scanning sequency obtains image next one pixel, step 2 is returned in redirect then; If handle the new images of then exporting behind the filtering mixed noise.

Claims (6)

1. method of filtering image mixed noise is characterized in that may further comprise the steps:
Step 1: the first pixel in original image that contains mixed noise of choosing input;
Step 2: differentiate this pixel and whether belong to impulsive noise,, then enter step 3, otherwise directly enter step 4 if this pixel belongs to impulsive noise;
Step 3: this pixel is carried out medium filtering handle, this impulsive noise is got rid of, directly enter step 6 then;
Step 4: the same group membership who determines this non-pulse noise pixel;
Step 5: by filtering filtering Gaussian noise on the same group;
Step 6: whether all pixels of check image dispose, and then obtain the next pixel of image if be untreated, and step 2 is returned in redirect then; If handle the new images of then exporting behind the filtering mixed noise;
Wherein, the described classification with the group membership of described step 4 further comprises:
Calculate this pixel grey scale second differnce in the horizontal direction;
Calculate this pixel grey scale second differnce in vertical direction;
Calculate the two-dimentional deviation branch of this pixel grey scale;
Calculate the direction of this pixel grey scale changing features minimum;
In this pixel filter window, obtain the same group membership that neighborhood territory pixel on this minimum direction constitutes this pixel.
2. method according to claim 1 is characterized in that whether this pixel of differentiation in the described step 2 belongs to impulsive noise and further comprise: differentiate impulsive noise by calculating the feature difference of this pixel on level, vertical or gradient direction.
3. method according to claim 1 is characterized in that whether this pixel of differentiation in the described step 2 belongs to impulsive noise and further comprise: judge according to the local energy feature of pixel whether this pixel is impulsive noise.
4. method according to claim 3, described local energy feature are judged and can be comprised:
With this test point pixel is the center, chooses the filter window of a n*n size;
Pixel grey scale average in the calculation of filtered window;
The local energy of calculating pixel;
The local energy average of choosing all pixels of image is used as the local energy threshold value;
If the local energy value of this pixel is judged that then it is an impulsive noise, otherwise is not impulsive noise greater than the local energy threshold value.
5. according to the described method of one of claim 1-4, it is characterized in that described step 4 further comprises: the distance of calculating earlier the characteristic vector of each neighborhood territory pixel and center pixel in filter window, then these distance values are carried out ascending order and arrange, adopt Fisher to differentiate at last and come definite member on the same group.
6. according to the described method of one of claim 1-4, it is characterized in that described step 4 further comprises: in filter window, directly will be classified as the same group membership of this pixel at the neighborhood territory pixel on this pixel characteristic gradient vertical direction.
CN2006100919045A 2006-06-10 2006-06-10 A method for filtering image mixed noise Expired - Fee Related CN101087365B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN2006100919045A CN101087365B (en) 2006-06-10 2006-06-10 A method for filtering image mixed noise

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN2006100919045A CN101087365B (en) 2006-06-10 2006-06-10 A method for filtering image mixed noise

Publications (2)

Publication Number Publication Date
CN101087365A CN101087365A (en) 2007-12-12
CN101087365B true CN101087365B (en) 2010-08-18

Family

ID=38938099

Family Applications (1)

Application Number Title Priority Date Filing Date
CN2006100919045A Expired - Fee Related CN101087365B (en) 2006-06-10 2006-06-10 A method for filtering image mixed noise

Country Status (1)

Country Link
CN (1) CN101087365B (en)

Families Citing this family (17)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
TWI456982B (en) 2010-03-30 2014-10-11 Realtek Semiconductor Corp Image processing device and spatial noise reducing method
CN101860667B (en) * 2010-05-06 2012-11-07 中国科学院西安光学精密机械研究所 Method for quickly removing mixed noise in image
CN102447817B (en) * 2010-09-30 2014-11-26 瑞昱半导体股份有限公司 Image processing device and space image noise eliminating method
CN105657216B (en) * 2010-11-19 2019-07-05 美国亚德诺半导体公司 Component for low smooth noise reduction filters
TWI488494B (en) 2011-04-28 2015-06-11 Altek Corp Method of multi-frame image noise reduction
CN102413338B (en) * 2011-11-21 2014-04-30 晶门科技(深圳)有限公司 Image processing method and device of spotted noise
CN104095640A (en) * 2013-04-03 2014-10-15 达尔生技股份有限公司 Oxyhemoglobin saturation detecting method and device
CN104240188B (en) * 2013-06-14 2017-09-12 华为技术有限公司 A kind of method and device for filtering out noise in pixel
CN105096263B (en) * 2014-05-22 2018-10-09 安凯(广州)微电子技术有限公司 image filtering method and device
CN104599252B (en) * 2015-01-29 2017-06-16 清华大学深圳研究生院 A kind of processing method for removing image mixed noise
CN105787902B (en) * 2016-03-22 2018-11-27 天津大学 Utilize the image denoising method of block sorting detection noise
KR102690986B1 (en) 2016-03-23 2024-08-02 삼성디스플레이 주식회사 Head-mounted device and display device
CN106373098B (en) * 2016-08-30 2019-04-23 天津大学 Method for suppressing random impulsive noise based on non-similar pixel statistics
CN109003247B (en) * 2018-07-26 2021-06-15 吉林大学 Method for removing color image mixed noise
CN109146816B (en) * 2018-08-22 2020-08-04 浙江大华技术股份有限公司 Image filtering method and device, electronic equipment and storage medium
CN109471014B (en) * 2018-10-30 2021-01-19 江苏赛诺格兰医疗科技有限公司 Detection signal simulation forming circuit and detector board card test platform
CN112861817A (en) * 2021-03-31 2021-05-28 国网上海市电力公司 Instrument noise image processing method

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1350747A (en) * 2000-01-13 2002-05-22 皇家菲利浦电子有限公司 Noise reduction
CN1383673A (en) * 2000-06-15 2002-12-04 皇家菲利浦电子有限公司 Noise filtering image sequence
CN1405734A (en) * 2002-10-28 2003-03-26 武汉大学 Method for reinforcing edge of medical picture

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1350747A (en) * 2000-01-13 2002-05-22 皇家菲利浦电子有限公司 Noise reduction
CN1383673A (en) * 2000-06-15 2002-12-04 皇家菲利浦电子有限公司 Noise filtering image sequence
CN1405734A (en) * 2002-10-28 2003-03-26 武汉大学 Method for reinforcing edge of medical picture

Also Published As

Publication number Publication date
CN101087365A (en) 2007-12-12

Similar Documents

Publication Publication Date Title
CN101087365B (en) A method for filtering image mixed noise
CN105787902B (en) Utilize the image denoising method of block sorting detection noise
US8300949B2 (en) Edge detection technique having improved feature visibility
CN106373098B (en) Method for suppressing random impulsive noise based on non-similar pixel statistics
CN101930597B (en) Mathematical morphology-based image edge detection method
CN102782706A (en) Text enhancement of a textual image undergoing optical character recognition
Shenbagavalli et al. Classification of soil textures based on laws features extracted from preprocessing images on sequential and random windows
US11703477B2 (en) Nondestructive testing method and device for detecting and distinguishing internal defect and external defect of wire rope
CN112085734A (en) GAN-based image restoration defect detection method
CN103337053A (en) Switching non-local total variation based filtering method for image polluted by salt and pepper noise
CN107742291A (en) A kind of the defects of photovoltaic glass detection method and device
CN104318586B (en) Adaptive morphological filtering-based motion blur direction estimation method and device
CN109146905A (en) For the CANNY operator edge detection algorithm of low-light level environment
CN106254723B (en) A kind of method of real-time monitoring video noise interference
CN101853496A (en) Hybrid image filtering method based on target scale
CN106780545A (en) A kind of Weak target energy accumulation Enhancement Method of combination time-space domain
Vignesh et al. Performance and Analysis of Edge detection using FPGA Implementation
Zhao et al. Analysis of image edge checking algorithms for the estimation of pear size
Niu et al. Application of CEM algorithm in the field of tunnel crack identification
CN106548460B (en) Laplce's detection and local similarity color image pulse noise filtering method
CN118229568B (en) Image edge de-pseudo noise reduction method and device, equipment and storage medium
Chudasama et al. Survey on Various Edge Detection Techniques on Noisy Images
CN109919853B (en) PIV image preprocessing method based on particle feature retrieval
CN114663451B (en) Edge detection method and system capable of processing noise and texture
CN114511591B (en) Track tracking method and device, electronic equipment and storage medium

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
C14 Grant of patent or utility model
GR01 Patent grant
TR01 Transfer of patent right
TR01 Transfer of patent right

Effective date of registration: 20171205

Address after: 065200 Langfang city of Hebei province Sanhe Yanjiao Development Zone Chaobai people two No. 18 Building 2 unit 303 room

Patentee after: Cheng Guihua

Address before: 518057, Nanshan District high tech Industrial Park, Guangdong province Shenzhen science and technology south road Zhongxing building A block 5

Patentee before: ZTE Corporation

CB03 Change of inventor or designer information
CB03 Change of inventor or designer information

Inventor after: Cheng Guihua

Inventor before: Cao Gang

Inventor before: Xu Lifeng

Inventor before: Luo Hongyu

CF01 Termination of patent right due to non-payment of annual fee
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20100818

Termination date: 20180610