CN103473742A - Novel method for suppressing noise in reflective off-axis digital holographic reconstructive image - Google Patents

Abstract

The invention provides a method for improving the quality of a reflective off-axis digital holographic reconstructive image, and discloses a novel method for suppressing noise in the reflective off-axis digital holographic reconstructive image. A zero-order image is suppressed through the Gaussian kernel function, and speckle noise is suppressed through improved non-local average filtering. The reproductive image is processed in a space domain, the zero-order image is effectively suppressed, and compared with a frequency domain method, the method is good in zero-order image suppressing effect, high in speed and suitable for dynamic measurement. Simulation is carried out through a computer with a 2.8GHz-processing-frequency dual-core CPU and a 2GB internal storage, the following simulation time is obtained: the time for suppressing a medal zero-order image through the Gaussian kernel function space domain suppression method is 0.61s, and the time through the frequency domain Gaussian high-pass filtering suppression method is 2.18s. The improved non-local average filtering is adopted, the speckle noise is suppressed, meanwhile, the detailed information in the image can be well kept, and the quality of the reproductive image is improved.

Description

Reflective off-axis digital holography reproduces the squelch new method in picture

Technical field

What the present invention relates to is a kind of method of improving reflective off-axis digital holography Reconstructed Image Quality amount, specific as follows: improving the Reconstructed Image Quality amount is for reproducing zero-order image and speckle noise in picture, for zero-order image, utilize the low-pass characteristic of gaussian kernel function, by at spatial domain digital hologram and gaussian kernel function, doing convolution algorithm, obtain the low-frequency information of hologram, recycle the low-frequency information that former digital hologram deducts acquisition, the processing of process said method can suppress zero-order image; For speckle noise, adopt the non-local mean filtering after improving to be suppressed.

Background technology

Digital Holography, along with the raising of CCD performance and the quickening of the operational speed of a computer, makes significant progress.Nowadays, this technology is widely used in scientific practice.For example, be successfully applied in the researchs such as living cells observation, transparent substance deformation, precise engineering survey.Although Digital Holography has the advantage of noncontact, high resolving power and real-time monitored, can be subject to the interference of zero-order image, conjugate image and speckle noise while reproducing, the details that makes to reproduce picture is smudgy, reduces the quality of reproducing picture.Therefore, in the reproduction picture, the inhibition of these noises is very important.

Off-axis holography can separate real image, conjugate image and zero-order image in space, make not interfere with each other between them, therefore is widely studied.But, when digital reproduction, conjugate image and zero-order image can form interference to reproducing picture.Especially zero-order image it occupied most diffraction energy.Meanwhile, rough due to body surface, cause the Random of scattered light, makes to reproduce as intensity fluctuation to change, and forms graininess and distribute, and speckle distributes or speckle noise, has a strong impact on the Reconstructed Image Quality amount.So during digital reproduction, the inhibition of zero-order image and speckle noise has extremely important impact to the quality of the quality of reproduction picture.

For the inhibition of zero-order image and speckle noise, mainly from two aspects, studied at present: a kind of method is based on optical technology.The inhibition of zero-order image has a step and two step phase-shift methods; The inhibition of speckle noise can be averaged by the hologram of recording the different wave length record, changes phase place or the transverse shifting CCD device of reference light.The method of optics can effectively suppress zero-order image and speckle noise, but increases the complicacy of multi-functional optical experimental device, is unfavorable for real-time measurement.The 2nd, adopt digital image processing techniques.The inhibition of zero-order image is generally, from spatial domain and frequency domain, hologram is carried out to the purpose that pre-service reaches the interference of inhibition zero-order image.Process and need positive inversefouriertransform on frequency domain, the processing time is slower; The inhibition of speckle noise generally adopts mean filter, wavelet threshold filtering and Lee filtering.Mean filter and wavelet threshold filtering are carried out filtering without distinction to the noise in image and detailed information, can cause a large amount of loss of information, cause image blurring unclear; Lee filtering is to derive on the basis of the property taken advantage of model, but sometimes the static state hypothesis of noise and the actual conditions of signal is not inconsistent, so effect unstable.

The purpose that this patent proposes is to improve the quality that reflective off-axis digital holography reproduces picture.This patent adopts gaussian kernel function to suppress zero-order image, and improved non-local mean filtering suppresses speckle noise.In spatial domain, to reproducing picture, processed, zero-order image has obtained effectively suppressing, and with frequency domain method, compares, and it is effective that this method suppresses zero-order image, and speed is applicable to kinetic measurement.Processing frequency at double-core CPU is 2.8GHz, and the Computer Simulation of internal memory 2GB obtains following simulation time: inhibition method in gaussian kernel function spatial domain suppresses medal zero-order image used time 0.61s, and frequency domain Gauss high-pass filtering suppresses method used time 2.18s.The detailed information of non-local mean filtering after adopt improving can keep image preferably when suppressing speckle noise in, improve the Reconstructed Image Quality amount.

Summary of the invention

The object of the invention is to improve the quality that reflective off-axis digital holography reproduces picture: at first utilize gaussian kernel function and digital hologram to do convolution algorithm, extract low-frequency information, recycle low-frequency information that former digital hologram deducts acquisition and just can obtain the digital hologram disturbed without zero-order image and reproduce and look like; Then on the basis on the reproduction picture disturbed without zero-order image, utilize improved non local filtering to carry out the inhibition of speckle noise, finally obtain high-quality reproduction picture.

Specific implementation step of the present invention is as follows:

1, the record of reflective off-axis digital holography figure: adopt the digital hologram experiment light path shown in Fig. 1, record digital hologram, as shown in Fig. 2 (a).

2, the inhibition of zero-order image: at first utilize the digital hologram convolution that gaussian kernel function and step 1 are recorded to extract low-frequency information, then utilize former digital hologram to deduct the low-frequency information of acquisition, finally obtain the digital hologram disturbed without zero-order image.On this basis, adopt Fresnel integral to reproduce the digital hologram disturbed without zero-order image, just can suppress to reproduce the zero-order image in picture, as shown in Fig. 2 (c).

3, the inhibition of speckle noise: on the basis of step 2, utilize improved non-local mean filtering to be suppressed speckle noise.Real image obtains the result as Fig. 3 (a) in design window function cut-away view 2 (c), from Fig. 3 (a), can find out image and have a large amount of speckle noises, makes image blurring unclearly, has affected the quality of image, so must carry out the inhibition of speckle noise.Non-local mean filtering (NLM) is proposed by people such as Buades, its ultimate principle is as follows: suppose a noisy image z={z (i) | i ∈ Ω }, Ω is image space, and the estimation of pixel i is wherein provided by formula (1), and it is other pixels in image

$\mathrm{ML}\left[z\right]\left(i\right)=\underset{j\∈\mathrm{\Ω}}{\mathrm{\Σ}}w(i,j)z\left(j\right)---\left(1\right)$

Weighted mean value, wherein w (i, j) is weighting coefficient, depends on the similarity of pixel i and pixel j, span is between 0 to 1, and ∑ _jw (i, j)=1.The similarity of pixel i and pixel j and gray-scale value vector z (N _i) and z (N _j) between similarity relevant, N _kthe Square Neighborhood of expression centered by pixel k.The calculating formula of w (i, j) is passed through one about Euclidean distance

decreasing function, a>0, mean the standard deviation of Gaussian function.The definition of w (i, j) as shown in formula (2),

$w(i,j)=\frac{1}{c\left(i\right)}\exp (-\frac{\left|\right|z\left(N_i\right)-z\left(N_j\right)|{|}_{2,a}^2}{h^2})---\left(2\right)$

Wherein c (i) is normaliztion constant, as shown in formula (3):

$c\left(i\right)=\underset{j}{\mathrm{\Σ}}\exp (-\frac{{\left|\right|z\left(N_i\right)-z\left(N_j\right)\left|\right|}_{2,a}^2}{h^2})---\left(3\right)$

Parameter h is the filtering degree, and it is controlling the attenuation degree of exponential function, more level and smooth after the larger filtering of h, otherwise level and smooth degree is less.

The accompanying drawing explanation

The reflective off-axis digital holography experiment of Fig. 1 light path, wherein 1-helium-neon laser; The 2-beam splitter; The 3-total reflective mirror; The 4-collimating apparatus; The 5-beam expander; The 6-imageing sensor; The 7-beam splitter; The 8-total reflective mirror; The 9-beam expander

The inhibition that the reflective off-axis digital holography of Fig. 2 reproduces zero-order image in picture

(a) reflective off-axis digital holography figure

(b) directly reproduce result

(c) Gauss's check figure suppresses zero-order image

(d) frequency domain Gauss Hi-pass filter

The inhibition that the reflective off-axis digital holography of Fig. 3 reproduces speckle noise in picture

(a) the reproduction picture that contains speckle noise

(b) non-local mean filtering

(c) improved non-local mean filtering

(d) averaging method

(e) wavelet thresholding method

(f) Lee filtering

Embodiment

Fig. 1 is digital hologram experiment light path, the incident laser of wavelength X=532nm is divided into two-beam by beam splitter S1, wherein, light wave by the S1 transmission is invested beam splitter S2 after total reflective mirror M1 reflects, collimates and expand, and through light wave formation and the optical axis z of S2 reflection, has the light wave arrival CCD of small angle to form reference light.Light wave by the S1 reflection forms the projection of illumination object light to object through reflection and after expanding, from the light wave of body surface scattering, through more than half anti-lens S2 arrival CCD, form object light.The copper medal of a 20km long-distance running of holding in Paris in 2000 during object being measured in experiment, diameter is about 60mm.When in experiment, CCD face battle array pixel count is 1024 * 1360 actual computation, number of samples is 1024 * 1024, pixel wide 4.65 μ m.Recording distance z ₀=1500mm, reproduce distance z _i=1500mm.

What the hologram of experimental record directly reproduced and reproduced after treatment the results are shown in Figure 2, Fig. 2 (a) this two spots occur at Zero-order diffractive spot left and right symmetrically, be due to reference light when the S2 during directive CCD, the S2 lower surface has faint object reflected light to see through S2 arrival CCD, the digital hologram that the reference light generated like this and faint object reflected light form after interfering, so just formed a pair of twin " object picture " but not zero-order image on the reproduction plane.Fig. 2 (c) is the reproduction result of Fig. 2 (a) after gaussian kernel function is processed, and can from Fig. 2 (c), find out that zero-order image has obtained inhibition.In image is processed, gaussian filtering generally has two kinds of implementations, and the one, by the sliding window convolution of discretize window, another kind passes through Fourier transform.Here adopt the sliding window convolution of discretize window, two-dimensional Gaussian function is first done the one dimension convolution to row, then row are done to the one dimension convolution.So just computation complexity can be dropped to O (2 * M * M * N) from O (M * M * N * N), M, N is respectively the window size of image and wave filter.In MATLAB, can utilize this software to carry function f special and produce the gaussian kernel function template, then adopt function imfilter, this template and digital hologram be done to two dimension cunning window convolution and carry out filtering.Fig. 2 (d) transforms to frequency domain through two-dimensional Fourier transform, carries out the result of reproducing after Gauss's high-pass filtering in frequency domain.As can be seen from Figure 2 good than the effect that adopts Fourier transform to suppress zero-order image at frequency domain by the sliding window convolution of discretize window in spatial domain.Processing frequency at double-core CPU is 2.8GHz, and the Computer Simulation of internal memory 2GB obtains following simulation time: inhibition method in gaussian kernel function spatial domain suppresses medal zero-order image used time 0.61s, and frequency domain Gauss high-pass filtering suppresses method used time 2.18s.

The search window of non-local mean filtering elects 7 * 7 as, and similar neighborhood window elects 3 * 3 as.The selection of parameter h affects the inhibition result of speckle noise, through test, selects 400 to be advisable, and when suppressing speckle noise, detailed information can farthest be protected like this.Non-local mean filtering is reproduced speckle noise suppressing result in picture as shown in Fig. 3 (b) to Fig. 3 (a), can find out that non-local mean filtering has very strong speckle noise suppressing power, but also damaged numerical portion in detailed information in image, especially image simultaneously, smudgy.For top problem.We have improved the distance of non-local mean filtering; parameter h is replaced by the average in the window of current pixel place; with the non-local mean filtering after improving to the filtering result of Fig. 3 (a) as shown in Fig. 3 (c); from Fig. 3 (c), can find out; not only the speckle noise in image 3 (a) has obtained inhibition; and detailed information also obtained good protection, numerical portion is also than clear much in Fig. 3 (b).We introduce speckle inhibition index SSI (Speckle Suppression Index) and edge keeps the image after index E PI (Edge Preserved Index) estimates denoising.The speckle inhibition index more bright speckle of novel suppresses more effective, and edge keeps the better of the larger explanation of index edge maintenance.Table 1 is that the speckle of each wave filter suppresses and edge keeps situation, as can be seen from Table 1, non-local mean filtering has best speckle restraint, the maintenance of Lee filtering edge is best, non-local mean filtering after improvement can suppress speckle noise effectively, also keep the edge information preferably, between non-local mean filtering and Lee filter.

The SSI of five kinds of filtering methods of table 1 and the comparison of EPI

The gaussian kernel function of this patent employing is slided zero-order image in window convolution inhibition off-axis digital holography reproduction picture in spatial domain, and zero-order image has obtained effectively suppressing, and with frequency domain method, compares, and this method inhibition zero-order image is effective, speed, applicable kinetic measurement.Utilize non-local mean filtering to suppress speckle noise, can find out that from the result of Fig. 3 and table 1 non-local mean filtering can suppress speckle noise preferably, but it is not strong to some details hold facility, the detailed information of non-local mean filtering after adopt improving can keep image preferably when suppressing speckle noise in, improve the Reconstructed Image Quality amount.

Claims (1)

1. one kind is improved the method that reflective off-axis digital holography reproduces the quality of picture, and specific implementation step of the present invention is as follows:

A, obtain reflective off-axis digital holography figure;

The inhibition of b, zero-order image: the digital hologram convolution of at first utilizing gaussian kernel function and step a to record is extracted low-frequency information, then utilize former digital hologram to deduct the low-frequency information of acquisition, finally obtain the digital hologram disturbed without zero-order image, on this basis, adopt Fresnel integral to reproduce the digital hologram disturbed without zero-order image, just can suppress to reproduce the zero-order image in picture;

The inhibition of c, speckle noise: on the basis of step b, utilize improved non-local mean filtering to be suppressed speckle noise, real image in design window function intercepting step b result, can find out in image from real image and have a large amount of speckle noises, make image blurring unclear, affected the quality of image, so must be suppressed speckle noise; Non-local mean filtering (NLM) is proposed by people such as Buades, and its ultimate principle is as follows: suppose a noisy image z={z (i) | i ∈ Ω }, Ω is image space, the estimation of pixel i wherein provided by formula (1),

$\mathrm{ML}\left[z\right]\left(i\right)=\underset{j\∈\mathrm{\Ω}}{\mathrm{\Σ}}w(i,j)z\left(j\right)---\left(1\right)$

I is the weighted mean value of other pixels in image, and w (i, j) is weighting coefficient, depends on the similarity of pixel i and pixel j, and span is between 0 to 1, and ∑ _jw (i, j)=1; The similarity of pixel i and pixel j and gray-scale value vector z (N _i) and z (N _j) between similarity relevant, N _kthe Square Neighborhood of expression centered by pixel k; The calculating formula of w (i, j) is passed through one about Euclidean distance

decreasing function, a>0, mean the standard deviation of Gaussian function, the definition of w (i, j) as shown in formula (2),

$w(i,j)=\frac{1}{c\left(i\right)}\exp (-\frac{\left|\right|z\left(N_i\right)-z\left(N_j\right)|{|}_{2,a}^2}{h^2})---\left(2\right)$

Wherein c (i) is normaliztion constant, as shown in formula (3):

$c\left(i\right)=\underset{j}{\mathrm{\Σ}}\exp (-\frac{{\left|\right|z\left(N_i\right)-z\left(N_j\right)\left|\right|}_{2,a}^2}{h^2})---\left(3\right)$

Parameter h is the filtering degree, and it is controlling the attenuation degree of exponential function, more level and smooth after the larger filtering of h, otherwise level and smooth degree is less; Here we have improved the distance of non-local mean filtering, parameter h is replaced by the average in the window of current pixel place, non-local mean filtering after improvement can suppress speckle noise effectively, and also keep the edge information preferably, further improve the Reconstructed Image Quality amount.

Images