CN102034227A

CN102034227A - Method for de-noising image

Info

Publication number: CN102034227A
Application number: CN 201010610864
Authority: CN
Inventors: 王德鹏; 范高生
Original assignee: Sichuan Jiuzhou Electric Group Co Ltd
Current assignee: Sichuan Jiuzhou Investment Holding Group Co.,Ltd.
Priority date: 2010-12-29
Filing date: 2010-12-29
Publication date: 2011-04-27
Anticipated expiration: 2030-12-29
Also published as: CN102034227B

Abstract

The invention relates to a method for processing a video image, and discloses a method for de-noising an image. The method comprises the following steps of: A, calculating the joint probability distribution P(I,J) of the image; B, determining the threshold range [LI,UI] of gray scale according to the P(I,J); C, ensuring that Wi,j is a N*N window, the coordinate of a central point of the window is (i,j), x is used for representing the central pixel point of the window Wi,j, and the window Wi,j moves to the top left corner of the image; and D, sequencing N*N pixel points in the Wi,j according to gray scale values, finding a gray scale value J corresponding to a median (N*N+1)/2 of the sequenced gray scale values, not processing the pixel point if J is in the threshold range [LI,UI] in the step B, and replacing the central point x of the window Wi,j with the gray scale value J if J is not in the threshold range [LI,UI]. The method has the advantage of solving the problems that image detail information is lost and the image blurs when pulse noise is removed by the conventional filtering method.

Description

A kind of method of image denoising

Technical field

The present invention relates to method of video image processing, relate in particular to a kind of method of image denoising.

Background technology

In intelligent transportation field widespread use video analysis and recognition technology, video analysis and identification (video analyzing and recognition) technology refers to use a computer and carry out computing and analysis by computational algorithm software from video, extract the useful information in the video, finish a technology of this information extraction and understanding, noise in the image can hinder the understanding of people to image, simultaneously also can hinder computer vision technique (CV, Computer Vision) accurately extracts useful information in the video, and the purpose of image denoising is exactly the noise of removing in the image, improve the awareness of people, improve image is done further useful information extraction and understood image; The video source of high-quality is provided for relevant Video Applications.Medium filtering denoising method of the prior art is a kind of nonlinear smoothing technology, and the gray-scale value of its each picture element is set to this intermediate value of putting all the picture element gray-scale values in certain neighborhood window.Taking out the odd number data by certain sample window from image sorts; Replacing data to be processed with the intermediate value after the ordering gets final product.But this method can cause losing of image detail information in removal impulsive noise, thereby image is thickened.And can in filtering noise, keep the detailed information of image effectively based on the impulsive noise filtering method that noise spot detects.The impulsive noise filtering method key that detects based on noise spot is: the one, and the detection of noise spot; The 2nd, to the filtering of noise spot.

Summary of the invention

What the present invention had overcome prior art can cause losing of image detail information removing impulsive noise, and therefore the problem that image thickens is necessary to provide a kind of method of image denoising.

A kind of image denoising method, it comprises following steps: steps A: the joint probability distribution P of computed image (I, J); Step B: (I J) determines the threshold range [LI, UI] of gray scale according to P; Step C:W _{I, j}Be the window of N*N, its center point coordinate is that (i, j), x is used for representing window W _{I, j}The center pixel point, described window W _{I, j}Move to the upper left corner of image; Step D: to W _{I, j}In N*N picture element sort according to the size of gray-scale value, find out intermediate value (N*N+1)/2 corresponding gray J of ordering gray-scale value, do not handle if J drops in the threshold range [LI, UI] among the step B this point; Otherwise window W _{I, j}Central point x replaces with gray-scale value J.

Described method also comprises step e: window W _{I, j}Move right one, if window W _{I, j}The low order end that moves to image then skips to next line; If window W _{I, j}If do not exceed the lower right corner of image, then go to step D and continue to carry out, otherwise whole denoising process finishes.

Described window W _{I, j}The upper left corner that moves to image is meant window W _{I, j}The left edge of a left side, coboundary and image overlap.Described method also comprises step F: the pictorial data after the output denoising.Described step B sets a threshold range for each grey level, and when whether the picture element of a certain grey level of judgement was noise spot in step C, one group of all corresponding with its grey level threshold range compared.Described window W _{I, j}Outwards continuation, the line number of continuation are (N-1)/2.

Compared with prior art, the invention has the beneficial effects as follows and avoided existing filtering method can cause losing of image detail information, the problem that image thickens removing impulsive noise.The detailed information that can in filtering noise, keep image based on the impulsive noise filtering method of noise spot detection effectively.

Description of drawings

Fig. 1 is for being step synoptic diagram of the present invention.

Embodiment

The present invention is further elaborated below in conjunction with accompanying drawing.The invention provides a kind of method of image denoising.It comprises following steps:

Steps A: the joint probability distribution P of computed image (I, J);

Step B: (I J) determines the threshold range [LI, UI] of gray scale according to P;

Step C:W _{I, j}Be the window of N*N, its center point coordinate is that (i, j), x is used for representing window W _{I, j}The center pixel point, described window W _{I, j}Move to the upper left corner of image;

Step D: to W _{I, j}In N*N picture element sort according to the size of gray-scale value, find out intermediate value (N*N+1)/2 corresponding gray J of ordering gray-scale value, do not handle if J drops in the threshold range [LI, UI] among the step B this point; Otherwise window W _{I, j}Central point x replaces with gray-scale value J.

The method of described image denoising also comprises step e: window W _{I, j}Move right one, if window W _{I, j}The low order end that moves to image then skips to next line; If window W _{I, j}If do not exceed the lower right corner of image, then go to step D and continue to carry out, otherwise whole denoising process finishes.

Described window W _{I, j}The upper left corner that moves to image is meant window W _{I, j}The left edge of a left side, coboundary and image overlap.The present invention also comprises step F: the pictorial data after the output denoising.

Preferably, all set a threshold range for each grey level among the step B, when whether the picture element of a certain grey level of judgement was noise spot in step C, one group of all corresponding with its grey level threshold range compared.By such mode, can reach the actual value of going back original image better.For example, 8 256 color shade figure have 256 groups of threshold values, and each group threshold value all is made up of lower limit LI and upper limit UI.Suppose that S is 8 256 color shade figure that contain the N*N of impulsive noise, Wi, j are the windows of N*N, its center point coordinate be (i, j), x is used for representing window Wi, the center pixel point of j, its gray-scale value is I.In order to judge whether x is noise spot, at first will be for S determines 256 groups of threshold ranges, for this this item technology defined a joint probability distribution P (I, J), I is window Wi, the gray-scale value of j central pixel point x, J is the value of this medium filtering. it is defined as follows:

Formula 1

Formula 2

I ∈ [0,255] in the formula, J ∈ [0,255], and med(medium filtering formula) (Wi, j) expression window Wi, the median-filtered result of j.In fact (I J) has reflected distribution relation between each gray level of image and its possible intermediate value result to P.According to P (I J) determines the threshold range [LI, UI] of each gray level, LI wherein, the definition of UI:

Formula 3

Formula 4

In the following formula, gray level is the medium filtering value of I among a presentation video S, and on behalf of moving window length, k be, is 3 herein, and t is a threshold value, and through experiment, when t got 2, denoising effect was best.

Further among technical scheme steps C and the step D, determined threshold range [LI, UI] afterwards through steps A and step B, to judge among step C and the step D that whether x is noise spot, at first will calculate x at corresponding window Wi, j, median-filtered result, if this result drops in the threshold value [LI, UI], judge that then this point is an image useful information point, if the result drops on threshold value [LI, UI] outside, judge that then this point is noise spot, then filtering.

Further among the technical scheme steps E, window Wi, j (then jump to next line if moved to the low order end of image) to the right and move one, if do not exceed the lower right corner of image, then go to step D and continue to carry out, otherwise whole filtering finishes.

Below with the gray-scale map of 8 512*512 as experimental subjects, added 10%, 20%, 30% low-light (level) noise respectively, Wi, j are the windows of a N*N, its center point coordinate is that (i, j), x is used for representing window Wi, the center pixel point of j.For the correct frontier point of handling image, at first tackle image and carry out expansion limit, border (this paper adopts image symmetrical to expand the limit method), and the line number on expansion limit depends on Wi, the size of N among the j, when N=3, Wi, j is the window of a 3*3, and the video in window border of 3*3 should be to extending out delegation; When N=5, Wi, j are the windows of a 5*5, and the window of 5*5 should be to extending out two row, and the line number that expands the limit in a word is (N-1)/2.Embodiments of the invention adopt the window of 3*3 should outwards extend out two row.

Expand the limit to set the value of the t in formula 3 and the formula 4 after finishing be 2 and carry out noise filtering according to steps A to the method flow of step F and handle also output image.Experimental results show that filter effect the best when t=2.

Practice shows that this noise filtering method can keep the detailed information of image effectively in filtering noise.More can embody the advantage of this method under the low-light (level) situation, compared with prior art low-light (level) video image details obtains extraordinary details reservation, for computer vision technique provides round-the-clock high-quality video image.

Claims

An image denoising method, it comprises following steps:

Steps A: the joint probability distribution P of computed image (I, J);

Step B: (I J) determines the threshold range [LI, UI] of gray scale according to P;

Step C:Wi, j are the windows of N*N, its center point coordinate be (i, j), x is used for representing window Wi, the center pixel point of j, described window Wi, j moves to the upper left corner of image;

Step D: to Wi, the picture element of the N*N among the j sorts according to the size of gray-scale value, finds out intermediate value (N*N+1)/2 corresponding gray J of ordering gray-scale value, if interior this point of threshold range [LI, UI] that J drops among the step B is not handled; Otherwise window Wi, j central point x replaces with gray-scale value J.
Image denoising as claimed in claim 1 method, it is characterized in that described method also comprises step e: window Wi, j move right one, if window Wi, the low order end that j moves to image then skips to next line; If window Wi, j if do not exceed the lower right corner of image, then go to step D and continue to carry out, otherwise whole denoising process finish.
Image denoising as claimed in claim 2 method, it is characterized in that described window Wi, the upper left corner that j moves to image is meant window Wi, the left side of the left side of j, coboundary and image, coboundary overlap.
Image denoising as claimed in claim 3 method, it is characterized in that described method also comprises step F: the pictorial data of output after the denoising.
Image denoising as claimed in claim 4 method, it is characterized in that described step B sets a threshold range for each grey level, when whether the picture element of a certain grey level of judgement was noise spot in step C, one group of all corresponding with its grey level threshold range compared.
Image denoising as claimed in claim 5 method, it is characterized in that described window Wi, the outside continuation of j, the line number of continuation is (N-1)/2.