CN101882305A - Method for enhancing image - Google Patents

Abstract

The embodiment of the invention discloses a method for enhancing an image, which comprises the following steps of: performing bi-dimensional empirical mode decomposition (BEMD) on an original image to obtain an intrinsic mode function (IMF) of the image; performing wavelet transformation on a high-frequency IMF in the IMF; performing wavelet adaptive de-noising on the high-frequency IMF after the wavelet transformation; performing wavelet inverse transformation on the high-frequency IMF after the wavelet adaptive de-noising; and synthesizing a low-frequency IMF in the IMF and the high-frequency IMF after the wavelet inverse transformation to obtain an enhanced image. By implementing the method for enhancing the image, the image can be enhanced while removing image noises, and useful information is highlighted.

Description

A kind of method of image enhancement processing

Technical field

The present invention relates to technical field of image processing, be specifically related to a kind of method of image enhancement processing.

Background technology

The figure image intensifying is an importance of Flame Image Process, it mainly adopts specific Enhancement Method to give prominence to some information in the image according to the ambiguity of concrete application scenarios and image, weaken simultaneously or eliminate irrelevant information, to reach the integral body of emphasizing image or the purpose of local feature.Along with the widespread use of image, the figure image intensifying has become one of urgent problem of key.

Algorithm for image enhancement is more at present, and image enhancement processing mode commonly used comprises greyscale transformation, histogram modification, image sharpening, noise remove, geometric distortion correction, frequency domain filtering, wavelet transformation etc.These algorithms are divided into based on the algorithm for image enhancement of time domain with based on the algorithm for image enhancement of frequency domain according to the difference of handling the space.Time domain refers to image itself, directly is treated to the basis based on the figure image intensifying of time domain with the pixel to image; Then revising the Fourier transform of image, the conversion coefficient (frequency components) of image is directly carried out computing, based on the algorithm for image enhancement of frequency domain then by the enhancing effect of Fourier inverse transformation with the acquisition image.The concrete technical scheme of these methods is as follows:

(1) greyscale transformation.Any r value in the intensity profile zone changes: s=T (r) wherein, T strengthens operation, T (r) changes the gray-level of image, method has that image is negated, enhancing contrast ratio and dynamic range compression etc.Compress mode commonly used is to adopt the conversion T of logarithmic form, s=clog (1+|r|).This method can increase the contrast of the integral body of image, but can't remove the denoising of image.

(2) histogram modification (histogram equalization).Histogram equalization (histogram equalization) thus be a kind ofly to realize that by histogram transformation grey scale mapping reaches the method for figure image intensifying purpose.Histogrammic horizontal ordinate is a gray level, and ordinate is the frequency that this gray level occurs; Histogram is the most basic statistical nature of image, and the intensity profile situation of original image is provided, and also we can say the integral body description that has provided the piece image gray-scale value.Basic thought: the histogram transformation of original graph is become equally distributed form, like this, just increased the dynamic range of grey scale pixel value, thereby reach the effect that strengthens the integral image contrast.But this method is not obvious to the removal of noise really.

(3) frequency domain filtering.HFS in the corresponding Fourier conversion of edge of image and noise, thus low-pass filter can smoothed image, remove out noise.The part of gradation of image generation fusion is corresponding with the high fdrequency component of frequency spectrum, so adopt the high frequency filter decay or suppress low frequency component, can carry out sharpening to image.But, because edge of image also is a high fdrequency component, so adopt low-pass filter will weaken image edge information.

In research and practice process to the method, the present inventor finds: image enchancing method mainly carries out in time domain or frequency domain, but the enhancement techniques that is combined into the basis with these two class methods is uncommon, and the method for using all is unilateral to the enhancing effect of image, has also lost more detailed information in filtering noise.At this problem, need a kind of details that in denoising, keeps image as much as possible.

Summary of the invention

The invention provides a kind of method of image enhancement processing, the method for traditional wavelet frequency domain filtering can be fused to the intrinsic mode function territory, can in denoising, keep the details of image as much as possible.

The invention provides a kind of method of image enhancement processing, comprising:

Original image is decomposed the intrinsic mode function IMF that BEMD obtains image by the two-dimensional empirical pattern;

High-frequency I MF among the IMF is carried out wavelet transformation;

The high-frequency I MF that carries out wavelet transformation is carried out the denoising of small echo self-adaptation;

The high-frequency I MF that carries out the denoising of small echo self-adaptation is carried out wavelet inverse transformation;

Low frequency IMF among the IMF and the high-frequency I MF that carries out wavelet inverse transformation are synthesized processing, the image after being enhanced.

Describedly original image decomposed the intrinsic mode function IMF that BEMD obtains image by the two-dimensional empirical pattern comprise:

A, r _I-1(x, y)=f (x, y) (i=1) wherein f (x y) is original image;

C, initialization: h _J-1(x, y)=r _I-1(x, y), j=1;

C, employing are determined h based on the method for the mathematical morphology reconstruct of geodesy operator _J-1(x, maximum value y) and minimum point adopt the cubic spline interpolation algorithm to obtain extreme value coenvelope up _J-1(x is y) with lower envelope down _J-1(x, y), and average envelope

D, from received image signal, deduct the average envelope surface:

h _j(x, y)=h _J-1(x, y)-m _J-1(x y), makes j=j+1;

E, calculating: $\mathrm{SD}=\underset{x,y}{\mathrm{\Σ}}\left[\frac{{\left|(h_j(x,y)-h_{j-1}(x,y))\right|}^2}{h_{j-1}^2(x,y)}\right]$

If SD＜sd (sd generally gets 0.2～0.3) then makes IMF _i(x, y)=h _j(x, y), otherwise repeating step d～e;

f、r _i(x，y)＝r _i-1(x，y)-IMF _i(x，y)；

G, i=i+1, repeating step b～f finishes until the IMF quantity (N) that obtains requiring.Can represent the image as this moment:

Wherein, IMF _i(x y) is two-dimentional IMF, R _N(x is y) corresponding to the residual error part after decomposing.

Because the embodiment of the invention adopts the two-dimensional empirical pattern to decompose and adaptive wavelet method, (Bidimensional Empirical Mode Decomposition, DEMD) it is based on data time domain local feature, adaptive time frequency analyzing tool to the decomposition of two-dimensional empirical pattern.It decomposes based on the local feature of signal, can be adaptively the data acquisition of any complexity be decomposed into a series of natural mode function limited, local arrowband (Intrinsic Mode Function by " sieve " process, IMF), it handles frequency information (distance between extreme point) in time domain.Decompose the high-frequency I MF obtain and decompose again by carrying out the two-dimensional empirical pattern, high-frequency I MF composition can be decomposed in the middle of the wavelet field like this, can be fully noise information be focused in the middle of the radio-frequency component, so that keep image detail as much as possible, and adopt improved wavelet thresholding method that its radio-frequency component is carried out denoising, therefore this invention can make the noise of removing image simultaneously with the figure image intensifying, and Useful Information is outstanding.

Description of drawings

In order to be illustrated more clearly in the embodiment of the invention or technical scheme of the prior art, to do to introduce simply to the accompanying drawing of required use in embodiment or the description of the Prior Art below, apparently, accompanying drawing in describing below only is some embodiments of the present invention, for those of ordinary skills, under the prerequisite of not paying creative work, can also obtain other accompanying drawing according to these accompanying drawings.

Fig. 1 is the process flow diagram of the image enhancement processing method in the embodiment of the invention;

Fig. 2 is that the two-dimensional empirical pattern is decomposed the process flow diagram of (BEMD);

Fig. 3 is based on the Lena image B EMD decomposing schematic representation of BEMD;

Fig. 4 is the small echo mapping function figure that single threshold is handled;

Fig. 5 is the small echo mapping function figure that dual threshold is handled;

Fig. 6 is the small echo mapping function figure of adaptive algorithm;

Fig. 7 is that the self-adaptation of small echo strengthens process flow diagram.

Embodiment

Below in conjunction with the accompanying drawing in the embodiment of the invention, the technical scheme in the embodiment of the invention is clearly and completely described, obviously, described embodiment only is the present invention's part embodiment, rather than whole embodiment.Based on the embodiment among the present invention, those of ordinary skills belong to the scope of protection of the invention not making all other embodiment that obtained under the creative work prerequisite.

The invention will be further described below in conjunction with accompanying drawing.

The embodiment of the invention provides a kind of and decomposes and the method for the figure image intensifying of adaptive wavelet based on the two-dimensional empirical pattern, the method of traditional wavelet frequency domain filtering the intrinsic mode function territory can be fused to, the details of image can be in denoising, kept as much as possible.Below be elaborated respectively.

Process flow diagram of the present invention comprises step as shown in Figure 1: (1) decomposes the intrinsic mode function (IMF) that BEMD obtains image by the two-dimensional empirical pattern; (2) the high frequency intrinsic mode function is carried out wavelet transformation; (3) realize adaptive denoising; (4) wavelet inverse transformation; (5) image after being enhanced.

Each step is specific as follows:

Step (1): the intrinsic mode function (IMF) that obtains image by BEMD

Propose a complete definition of intrinsic mode function in the existing technology, and should satisfy following two conditions:

(1) in whole data sequence, the number of extreme point and the number of zero crossing must equate, or differ one at most.Promptly

(N _z-1)≤N _e≤(N _z+1)

N wherein _eRepresent the extreme point number, comprise maximum point and minimum point; N _zRepresent the zero crossing number.

(2) on the time point in office, the mean value of the lower envelope that coenvelope that is formed by the local maximum point of signal and local minizing point form is zero.Promptly

$\frac{[f_{\max }\left(t_i\right)+f_{\min }\left(t_i\right)}{2}=0,t_i\∈[t_a,t_b]$

Wherein, f _Max(t) representative is by the definite coenvelope of maximum point; f _Min(t) representative is by the definite lower envelope of minimum point.

When two-dimensional image data is carried out the BEMD decomposition, we must make following hypothesis earlier: the 2-D data plane comprises a maximum point and a minimum point at least, or whole two dimensional surface does not have extreme point but a maximum point and a minimum point can occur after carrying out single order or multistage derivative operation; Characteristic dimension distance definition between the extreme point.

As shown in Figure 2, the basic thought of two-dimensional empirical pattern decomposition (BEMD) method is as follows:

A, r _I-1(x, y)=f (x, y) (i=1) wherein f (x y) is original image;

B, initialization:

h _j-1(x，y)＝r _i-1(x，y)，j＝1；

C, employing are determined h based on the method for the mathematical morphology reconstruct of geodesy operator _J-1(x, maximum value y) and minimum point adopt the cubic spline interpolation algorithm to obtain extreme value coenvelope up _J-1(x is y) with lower envelope down _J-1(x, y), and average envelope

D, from received image signal, deduct the average envelope surface:

h _j(x, y)=h _J-i(x, y)-m _J-1(x y), makes j=j+1;

E, calculating:

$\mathrm{SD}=\underset{x,y}{\mathrm{\Σ}}\left[\frac{{\left|(h_j(x,y)-h_{j-1}(x,y))\right|}^2}{h_{j-1}^2(x,y)}\right]$

If SD＜sd (sd generally gets 0.2～0.3) then makes IMF _i(x, y)=h _j(x, y), otherwise repeating step d～e;

f、r _i(x，y)＝r _i-1(x，y)-IMF _i(x，y)；

G, i=i+1, repeating step b～f finishes until the IMF quantity (N) that obtains requiring.Can represent the image as this moment:

$f(x,y)=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{\mathrm{IMF}}_i(x,y)+R_N(x,y)$

Wherein, IMF _i(x y) is two-dimentional IMF, R _N(x is y) corresponding to the residual error part after decomposing.

Shown in Figure 2 is 4 IMF components and residual error remainder that obtains after image decomposes through BEMD.

Step (2): the high frequency intrinsic mode function is carried out wavelet transformation

With 4 IMF components that obtain in the step (1) are high fdrequency components, and the residual error remainder then is a low frequency component.IMF is carried out 5 layers of wavelet decomposition, obtain the wavelet coefficient of each IMF component respectively.

Shown in Figure 3 is that IMF is through the synoptic diagram behind the wavelet transformation.

Step (3): adaptive denoising

Afterwards wavelet coefficient is carried out processing and amplifying through step (2).Problem is, if original image contains noise, so many wavelet coefficients, particularly the wavelet coefficient on the high-resolution yardstick is caused by noise, then when carrying out inverse wavelet transform (normally final step), any amplification to these wavelet coefficients all can be reflected in the enhancing to noise.

Single threshold strengthen to adopt is a kind of simple but mapping function of effectively processing wavelet coefficient, as shown in Figure 3.By analysis chart 2 as can be known this mapping function be expressed as:

This single G is gain, W _InBe the wavelet coefficient values of input, W _OutIt is the wavelet coefficient values of output.Use this mapping function can amplify noise.Noise in the middle of the image shows as less wavelet coefficient in wavelet transformed domain, and this single mapping function all amplifies by maximum gain the wavelet coefficient that absolute value is lower than T, has therefore caused the amplification of noise inevitably.

It is that what to adopt is a kind of mapping function that uses the processing wavelet coefficient of dual threshold that dual threshold strengthens, as shown in Figure 4.The wavelet coefficient that all absolute values are lower than less threshold value T1 in this function all is suppressed (just being set to 0); Absolute value is greater than T1 but have identical gain G less than the wavelet coefficient of higher thresholds T2, and for the wavelet coefficient of absolute value greater than T2, its gain reduces along with the increase of wavelet coefficient values.The mapping function of this form can be defined as follows:

$W_{\mathrm{out}}=\left\{\begin{array}{c}{W}_{\mathrm{in}}+T_2^{*}(G-1)-T_1^{*}G,T_2<W_{\mathrm{in}};\\ G*(W_{\mathrm{in}}-T_1),T_1<W_{\mathrm{in}}\&le;T_2;\\ 0,-T_1\&le;W_{\mathrm{in}}\&le;T_1;\\ G*(W_{\mathrm{in}}+T_1),-T_2\&le;W_{\mathrm{in}}<-T_1;\\ W_{\mathrm{in}}-T_2^{*}(G-1)+T_1^{*}G,W_{\mathrm{in}}<T_2.\end{array}\right.$

Analysis chart 3 not only can be eliminated noise if suppress the wavelet coefficient of very little value as can be known, also can blur or damage detail signal simultaneously.Use the minor detail of the mapping function meeting failure pattern picture of this form, this just disagrees with the original intention of carrying out enhancement process.

In order to improve wavelet coefficient values arbitrarily again when suppressing noise, we need a kind ofly can distinguish which coefficient is by the caused by noise method in the wavelet coefficient, self-adapting enhancement method, and its mapping function is as shown in Figure 5.This method needn't only rely on the size of wavelet coefficient values and handle, and promptly will the following wavelet coefficient values of all Fujian values all not suppress from day ground.We can utilize this feature of correlativity of wavelet coefficients between two neighboring scales on the same space position.In fact, the correlativity of the wavelet coefficient that caused by noise of the same position place between adjacent yardstick is all very little usually.Therefore, we just can utilize the correlativity of wavelet coefficient under different scale come differentiate between images in the middle of which wavelet coefficient cause that by noise which is caused by signal itself again.

At this problem, the present invention judges that by the confidence level of calculating noise wavelet coefficient is that noise causes or signal causes.If known yardstick is n and adjacent yardstick is the wavelet coefficient on the n+1, our needs multiply each other two wavelet coefficient values of each pixel seriatim so, just can obtain correlation on each location of pixels:

Corr _n＝W _n*W _n+1

Calculate the general power of n layer correlation and the general power of wavelet coefficient:

P _Corr＝∑Corr _n*Corr _n

P _W＝∑W _n*W _n

And then can get the correlation of following form:

$\mathrm{Corr}{2}_n={\mathrm{Corr}}_n*\sqrt{P_W/P_{\mathrm{Corr}}}$

Can obtain now and minimum absolute value in the correlation:

Corr2 _min＝min{|Corr2 _n|}

This value can be used for the mark noise: all Corr2 _n, absolute value equals Corr2 _MinThe position we think that wavelet coefficient is caused by noise, its noise confidence level E is set to 1; To Corr2 _nAbsolute value compares W _nWe think that wavelet coefficient is caused by signal the big position of absolute value, and it is 0 that E is set in these positions; And for Corr2 _nAbsolute value compares W _nAbsolute value is little but compare Corr2 _MinBig position, we are provided with E is value between one 0 to 1, this value is that W produces, absolute value and Corr2 _nThe difference and the W of absolute value _nAbsolute value and Corr2 _MinThe ratio of difference, that is:

$E=\frac{\mathrm{abs}\left(W_n\right)-\mathrm{abs}\left({\mathrm{Corr}2}_n\right)}{\mathrm{abs}\left(W_n\right)-{\mathrm{Corr}2}_{\min }}$

Cause by noise when E shows wavelet coefficient, should adopt among Fig. 5 that thinner defined mapping function in two solid lines so.On the other hand, be not by caused by noise if E shows wavelet coefficient, so at this moment should select the pairing mapping function of that solid line thicker among Fig. 5.E can not judge wavelet coefficient be on earth by noise or situation about causing by signal under, gain gets 0 to G _MaxA middle value.Particularly: be made as 0 if E greater than the upper limit 0.9 (experiment value), then gains; If E is less than lower limit 0.75 (experiment value), then gain is made as maximal value G _Max, between 0.9 and 0.75, then gain linearity changes as if E.

Shown in Fig. 6 process flow diagram, will carry out 5 layers of wavelet decomposition to each IMF component among the present invention, the 1st layer is used dual threshold to strengthen, and the 2nd, 3 layer is used self-adaptation to strengthen, and the 4th, 5 layer is used single threshold to strengthen.After handling respectively, obtain new wavelet coefficient.

Step (4): carry out wavelet inverse transformation

Use wavelet inverse transformation to obtain new IMF component IMF ' respectively through the IMF after step (3) processing.

Step (5): the image after being enhanced

Through the R after step (4) IMF ' afterwards and the BEMD _N(x, y) image of summation after being enhanced is promptly:

Because the embodiment of the invention adopts the two-dimensional empirical pattern to decompose and adaptive wavelet method, (Bidimensional Empirical Mode Decomposition, DEMD) it is based on data time domain local feature, adaptive time frequency analyzing tool to the decomposition of two-dimensional empirical pattern.It decomposes based on the local feature of signal, can be adaptively the data acquisition of any complexity be decomposed into a series of natural mode function limited, local arrowband (Intrinsic Mode Function by " sieve " process, IMF), it handles frequency information (distance between extreme point) in time domain.Decompose the high-frequency I MF obtain and decompose again by carrying out the two-dimensional empirical pattern, high-frequency I MF composition can be decomposed in the middle of the wavelet field like this, can be fully noise information be focused in the middle of the radio-frequency component, so that keep image detail as much as possible, and adopt improved wavelet thresholding method that its radio-frequency component is carried out denoising, therefore this invention can make the noise of removing image simultaneously with the figure image intensifying, and Useful Information is outstanding.

Need to prove, contents such as the information interaction between said apparatus and intrasystem each unit, implementation since with the inventive method embodiment based on same design, particular content can repeat no more referring to the narration among the inventive method embodiment herein.

One of ordinary skill in the art will appreciate that all or part of step in the whole bag of tricks of the foregoing description is to instruct relevant hardware to finish by program, this program can be stored in the computer-readable recording medium, storage medium can comprise: ROM (read-only memory) (ROM, Read Only Memory), random access memory (RAM, Random Access Memory), disk or CD etc.

More than the embodiment that the embodiment of the invention provided is described in detail, used specific case herein principle of the present invention and embodiment are set forth, the explanation of above embodiment just is used for helping to understand method of the present invention and core concept thereof; Simultaneously, for one of ordinary skill in the art, according to thought of the present invention, the part that all can change in specific embodiments and applications, in sum, this description should not be construed as limitation of the present invention.

Claims (2)

1. the method for an image enhancement processing is characterized in that, comprising:

Original image is decomposed the intrinsic mode function IMF that BEMD obtains image by the two-dimensional empirical pattern;

High-frequency I MF among the IMF is carried out wavelet transformation;

The high-frequency I MF that carries out wavelet transformation is carried out the denoising of small echo self-adaptation;

The high-frequency I MF that carries out the denoising of small echo self-adaptation is carried out wavelet inverse transformation;

Low frequency IMF among the IMF and the high-frequency I MF that carries out wavelet inverse transformation are synthesized processing, the image after being enhanced.

2. the method for claim 1 is characterized in that, describedly original image is decomposed the intrinsic mode function IMF that BEMD obtains image by the two-dimensional empirical pattern comprises:

A, r _I-1(x, y)=f (x, y) (i=1) wherein f (x y) is original image;

B, initialization: h _J-1(x, y)=r _I-1(x, y), j=1;

C, employing are determined h based on the method for the mathematical morphology reconstruct of geodesy operator _J-1(x, maximum value y) and minimum point adopt the cubic spline interpolation algorithm to obtain extreme value coenvelope up _J-1(x is y) with lower envelope down _J-1(x, y), and average envelope

D, from received image signal, deduct the average envelope surface:

h _j(x, y)=h _J-i(x, y)-m _J-1(x y), makes j=j+1;

E, calculating: $\mathrm{SD}=\underset{x,y}{\mathrm{\&Sigma;}}\left[\frac{{\left|(h_j(x,y)-h_{j-1}(x,y))\right|}^2}{h_{j-1}^2(x,y)}\right]$

If SD＜sd (sd generally gets 0.2～0.3) then makes IMF _i(x, y)=h _j(x, y), otherwise repeating step d～e;

f、r _i(x，y)＝r _i-1(x，y)-IMF _i(x，y)；

G, i=i+1, repeating step b～f finishes until the IMF quantity (N) that obtains requiring.Can represent the image as this moment:

Wherein, IMF _i(x y) is two-dimentional IMF, R _N(x is y) corresponding to the residual error part after decomposing.

Images