Image noise suppression method and system for laser speckle

Abstract

The application discloses a kind of image noise suppression method and system for laser speckle, it is related to image noise suppression technical field, first, two laser images of same position are acquired, and two-dimensional discrete wavelet transform is carried out, and high, low frequency image is obtained;High, low frequency image is fused, and high-frequency fusion image and low-frequency fusion image are obtained;Optimized non-local mean filtering algorithm is used to filter high-frequency fusion image, and filtered high-frequency fusion image is obtained;By wavelet inverse transform, low-frequency fusion image and filtered high-frequency fusion image are restored, and restoration image is obtained;Optimized non-local mean filtering algorithm is used to filter restoration image, and final result image is obtained.The application is aimed at laser speckle high-frequency noise, introduces wavelet transform and improved optimized non-local mean filtering combination, not only eliminates high-frequency speckle noise in laser image, but also completely retains low-frequency information part, and the obtained image effect is best.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image noise suppression, and more particularly to an image noise suppression method and system for laser speckle. BACKGROUND

[0002] Laser imaging technology has the advantages of non-contact, wide area and continuous long-term detection, so MEMS laser three-dimensional reconstruction technology has important significance in precision detection, and has been widely used in industrial detection, security monitoring, intelligent transportation, medical treatment and other fields; the technology can improve the detection distance, obtain target detail features, identify target camouflage and distinguish decoys.

[0003] In the process of measuring the grating structure light image, due to the multiple interference of external factors and experimental equipment, the noise existing in the image will affect the image quality, reduce the resolution of the target object, and then affect the accuracy of the subsequent phase unwrapping phase information, and seriously affect the accuracy of three-dimensional measurement. Image filtering technology is to attenuate or eliminate the adverse effects of noise on images, improve image quality, and improve image contrast and resolution.

[0004] Image filtering technology is usually divided into methods based on spatial domain, methods based on transform domain and methods based on neural networks. Spatial domain methods include histogram equalization, image averaging, image sharpening using edge detection and morphological operators, and nonlinear median filtering. The method based on transform domain involves the transformation of the domain, such as Fourier transform, and the frequency transform method is beneficial to extract some image features that cannot be obtained from the spatial domain. Due to the particularity of MEMS structured light technology, it is difficult to obtain a noise-free original image as a training set, so the noise reduction algorithm based on neural networks is not commonly used. Therefore, when eliminating the noise of laser images, the noise reduction algorithm of spatial domain or transform domain is usually selected. The non-local mean (NLM) filtering algorithm proposed by Buades et al. uses the redundant information in the image to make the noise reduction effect of the image better. This algorithm can eliminate the noise in the laser image while eliminating the main information part in the image, and can lose some feature information while eliminating the noise.

[0005] In the document with application number "202111527527.6", an image denoising method and device based on MEMS stripe structured light are given, which includes improving the weight calculation function in the traditional non-local mean filtering algorithm, thereby improving the precision of the traditional non-local mean denoising algorithm. The invention optimizes the weight function to eliminate the speckle noise in the laser image, but at the same time, a part of the feature information is also eliminated, and the low-frequency information of the image is not completely preserved.

[0006] Therefore, how to solve the problem of losing low-frequency feature information while eliminating noise in the prior art is an urgent problem for those skilled in the art. SUMMARY

[0007] Therefore, the application provides an image noise suppression method and system for laser speckle, which can not only eliminate speckle high-frequency noise in a laser image, but also retain as much image information as possible.

[0008] To achieve the above object, the application provides the following technical scheme.

[0009] An image noise suppression method for laser speckle comprises the following steps.

[0010] Step 1: obtaining two laser images P1 and P2 at the same position;

[0011] Step 2: performing two-dimensional discrete wavelet transform on the two laser images P1 and P2 to obtain high-frequency images HH1, HL1 and LH1 of the laser image P1, low-frequency image LL1 of the laser image P1, high-frequency images HH2, HL2 and LH2 of the laser image P2, and low-frequency image LL2 of the laser image P2;

[0012] Step 3: fusing the high-frequency images and the low-frequency images of the laser images P1 and P2 to obtain high-frequency fused images HH', HL' and LH' and low-frequency fused image LL';

[0013] Step 4: using an optimized non-local mean filter algorithm to filter the high-frequency fused images HH', HL' and LH' respectively to obtain filtered high-frequency fused images HH'', HL'' and LH'';

[0014] Step 5: restoring the low-frequency fused image LL' and the filtered high-frequency fused images HH'', HL'' and LH'' by wavelet inverse transform to obtain a restored image P';

[0015] Step 6: using the optimized non-local mean filter algorithm to filter the restored image P' to obtain a final result image.

[0016] Optionally, in step 3, the specific method for fusing the high-frequency images and the low-frequency images of the laser images P1 and P2 is as follows.

[0017] For the low-frequency images: the low-frequency image LL1 of the laser image P1 and the low-frequency image LL2 of the laser image P2 are fused by using an average value method, LL'(x, y) = [LL1(x, y) + LL2(x, y)] / 2;

[0018] For the high frequency image: using the weight value fusion method, the high frequency image HH1 of the laser image P1 and the high frequency image HH2 of the laser image P2 are fused, the high frequency image HL1 of the laser image P1 and the high frequency image HL2 of the laser image P2 are fused, and the high frequency image LH1 of the laser image P1 and the high frequency image LH2 of the laser image P2 are fused.

[0019] Optionally, the specific process of fusing the high frequency image using the weight value fusion method is as follows:

[0020] The weight values ξ1(x,y) and ξ2(x,y) of the two high frequency images are calculated respectively.

[0021] The fusion coefficient matrix of the two high frequency images is solved according to the weight values ξ1(x,y) and ξ2(x,y) of the two high frequency images.

[0022] The two high frequency images are fused using the fusion coefficient matrix.

[0023] Optionally, the calculation formula of the weight value is ξ(x,y)=|LH u (x,y)-LH η (x,y)|, wherein, LH u (x,y) is the result of the mean filtering of the high frequency image, and LH η (x,y) is the result of the median filtering of the high frequency image.

[0024] Optionally, the fusion coefficient matrix is represented as:

[0025]

[0026]

[0027] Optionally, in the step 4, the optimized non-local mean filtering algorithm is improved on the weight function on the basis of the traditional non-local mean filtering algorithm, and the improved weight function is represented as w'(x,y):

[0028]

[0029] wherein, n is a weight control parameter; |N(x)-N(y)|| 2 represents the Euclidean distance between the neighborhoods; Z(x) is a normalization coefficient, and h is a smoothing parameter.

[0030] Optionally, the Euclidean distance between the neighborhoods is represented as:

[0031]

[0032] The normalization coefficient is represented as:

[0033]

[0034] wherein, ds is a neighborhood window radius; d = 2*ds + 1; S t is an image of assumed pixel difference; x1, x2 are x coordinates of different pixels.

[0035] An image noise suppression system for laser speckle, comprising:

[0036] An image acquisition module is configured to acquire two laser images P1 and P2 at the same position.

[0037] An image decomposition module is configured to perform two-dimensional discrete wavelet transform on the two laser images P1 and P2 to obtain high-frequency images HH1, HL1 and LH1 of the laser image P1, a low-frequency image LL1 of the laser image P1, high-frequency images HH2, HL2 and LH2 of the laser image P2, and a low-frequency image LL2 of the laser image P2.

[0038] An image fusion module is configured to fuse the high-frequency images and the low-frequency images of the laser images P1 and P2 to obtain high-frequency fused images HH', HL' and LH' and a low-frequency fused image LL'.

[0039] A high-frequency fused image filtering module is configured to use an optimized non-local mean filtering algorithm to filter the high-frequency fused images HH', HL' and LH' respectively to obtain filtered high-frequency fused images HH'', HL'' and LH''.

[0040] An image restoration module is configured to restore and recover the low-frequency fused image LL' and the filtered high-frequency fused images HH'', HL'' and LH'' through inverse wavelet transform to obtain a restored image P'.

[0041] A secondary filtering module is configured to use the optimized non-local mean filtering algorithm to filter the restored image P' to obtain a final result image.

[0042] According to the technical solution described above, the present application provides an image noise suppression method and system for laser speckle, which has the following advantages compared with the prior art:

[0043] The Non-Local Means (NLM) filtering algorithm solves the laser image noise, the algorithm still gives a certain weight even if the two pixel points are very different; and the function is too sensitive when the pixel points are close, and drops too fast. In addition, the NLM algorithm does not specifically eliminate the speckle noise existing in the high frequency component. The wavelet transform is introduced, the wavelet transform not only has good reconstruction ability, and it is easy to obtain the structure information and detail information in the image. When the high frequency information of the two laser images is fused by wavelet transform, the idea of visual saliency detection is introduced, and the high frequency redundant information of the two images is combined. When filtering the high frequency speckle noise, the weight function in the non-local mean filtering algorithm is improved and optimized, so that it meets: with the change of the distance between the two pixel points, the weight value is well adapted to the distance change, when the two pixel points are too different, the calculated weight value is best close to 0, so that it can prevent introducing too much noise error.

[0044] In summary, the present application is aimed at laser speckle high frequency noise, wavelet transform and improved non-local mean filtering are combined, not only the high frequency speckle noise in the laser image is eliminated, but also the low frequency information part is completely reserved, and the obtained image effect is best. BRIEF DESCRIPTION OF DRAWINGS

[0045] In order to more clearly illustrate the technical solutions of the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiment or prior art description, obviously, the drawings in the following description are only embodiments of the present application, and for those skilled in the art, other drawings can be obtained without creative labor on the basis of the provided drawings.

[0046] Figure 1 It is a schematic diagram of two-dimensional discrete wavelet transform (DWT);

[0047] Figure 2 It is an integral graph;

[0048] Figure 3 It is a principle diagram of the image noise suppression method for laser speckle of the present application;

[0049] Figure 4 It is a curve diagram of the original weight function and the optimized weight function;

[0050] Fig. 5(a) is the original laser image in example 3;

[0051] Fig. 5(b) is the image obtained by non-local mean filtering in example 3;

[0052] Fig. 5(c) is the image obtained by the method of the present application in example 3;

[0053] Figure 6(a) is a local enlarged view of the original laser image in Example 3;

[0054] Figure 6(b) is a local enlarged view of the image obtained by non-local mean filtering in Example 3;

[0055] Figure 6(c) is a local enlarged view of the image obtained by the method of the present application in Example 3. DETAILED DESCRIPTION

[0056] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0057] Example 1

[0058] The embodiments of the present application disclose a method for suppressing image noise of laser speckle. The MEMS laser coherent imaging system is used to collect a calibration board as a target image, and filtering is performed. Referring to Figure 3 , the method specifically comprises the following steps:

[0059] Step 1, acquiring two laser images P1 and P2 of a calibration board with unchanged relative positions;

[0060] Step 2, using Discrete Wavelet Transform (DWT) to decompose a two-dimensional image into high and low frequency information, which is represented by a formula:

[0061]

[0062]

[0063]

[0064]

[0065] wherein, is a two-dimensional scale function, which measures the change of gray scale along different directions in the image through a separable "direction-sensitive" wavelet: Ψ H , Ψ V and Ψ D are along the horizontal edge, vertical edge and diagonal direction, respectively.

[0066] Referring to Figure 1The embodiment uses two-dimensional discrete wavelet transform to process two laser images P1 and P2, 2↓ represents downsampling, and high-frequency filters (h high ) and low-frequency filters (h low ) are used to obtain high-frequency images HH1, HL1 and LH1 of the laser image P1, low-frequency image LL1 of the laser image P1, high-frequency images HH2, HL2 and LH2 of the laser image P2, and low-frequency image LL2 of the laser image P2.

[0067] Step 3, fusing the high-frequency images and low-frequency images of the laser images P1 and P2 to obtain high-frequency fused images HH', HL' and LH', and low-frequency fused image LL'.

[0068] For the low-frequency images, the low-frequency image LL1 of the laser image P1 and the low-frequency image LL2 of the laser image P2 are fused by using the average value method, LL'(x, y) = [LL1(x, y) + LL2(x, y)] / 2.

[0069] For the high-frequency images, according to the theoretical basis of visual saliency detection, the high-frequency image HH1 of the laser image P1 and the high-frequency image HH2 of the laser image P2 are fused by using the weight value fusion method to obtain the high-frequency fused image HH'; the high-frequency image HL1 of the laser image P1 and the high-frequency image HL2 of the laser image P2 are fused to obtain the high-frequency fused image HL'; and the high-frequency image LH1 of the laser image P1 and the high-frequency image LH2 of the laser image P2 are fused to obtain the high-frequency fused image LH'.

[0070] For example, for two high-frequency images p1 and p2, the fusion process by using the weight value fusion method is as follows:

[0071] Step (1), the weight values ξ1(x, y) and ξ2(x, y) of the two high-frequency images p1 and p2 are calculated respectively; the calculation formula of the weight value is ξ(x, y) = |LH u (x, y) - LH η (x, y)|, wherein LH u (x, y) is the result of mean filtering of the high-frequency image, and LH η (x, y) is the result of median filtering of the high-frequency image.

[0072] Step (2), according to the weight values ξ1(x, y) and ξ2(x, y) of the two high-frequency images p1 and p2, the fusion coefficient matrix of the two high-frequency images p1 and p2 is solved:

[0073]

[0074]

[0075] Step (3), using the fusion coefficient matrix, fusing two high-frequency images p1 and p2.

[0076] Step 4, the speckle noise in the image obtained by the laser imaging system exists in the high-frequency component of the image, and further affects the detail information of the laser image, therefore, the application uses an improved non-local means filtering algorithm (INLM) to filter the high-frequency fused images HH', HL' and LH', respectively, to obtain the filtered high-frequency fused images HH'', HL'' and LH''.

[0077] Let the decomposed high-frequency noise image be v, and the filtered image be u. The pixel value at x in u is represented by the formula:

[0078]

[0079] where w(x, y) is the similarity between pixel points x and y, and its value is represented by the Gaussian weighted Euclidean distance between the neighborhoods N(x) and N(y) centered at x and y:

[0080]

[0081] The w(x, y) weight function is optimized so that when the distance between two pixel points is close, a high weight value is assigned, and when the distance is far, a low weight value is assigned. At the same time, the weight value decreases rapidly within a certain range, so that the weight value assigned to different distance pixel values is more accurate. The optimized weight function is represented by w'(x, y):

[0082]

[0083] where n is the weight control parameter, and its value affects the ideal degree of the weight function; h is the smoothing parameter, which controls the strength of the noise reduction effect. See Figure 4 , which is a schematic diagram of the original weight function and the optimized weight function.

[0084] When n is too large, the calculation of the weight value only considers the pixel values of the nearby pixels, and does not consider the pixel values of the middle distance pixels. Therefore, selecting an appropriate n value can make different pixels obtain the best weight value and improve the image filtering effect. When n is selected as 8, the weight function considers two pixel points of middle distance and assigns them different weights.

[0085] In solving the inter-domain similarity, the calculation complexity of the similarity between neighborhoods can be optimized by the idea of integral graph to improve the running speed of the algorithm, such as Figure 2 is the idea according to the integral graph. The Euclidean distance between the two neighborhoods and the normalization parameter are represented as:

[0086]

[0087]

[0088] The Euclidean distance between the neighborhoods is expressed as:

[0089]

[0090] The normalized coefficient is expressed as:

[0091]

[0092] wherein, ds is the neighborhood window radius; d = 2*ds + 1; S t is an image of a hypothetical pixel difference; x1, x2 are x coordinates of different pixels.

[0093] Step 5, by wavelet inverse transform, the low-frequency fusion image LL' and the filtered high-frequency fusion image HH'', HL'', LH'' are restored to obtain a restored image P';

[0094] Step 6, using an optimized non-local mean filter algorithm, the restored image P' is filtered to obtain a final result image.

[0095] Embodiment 2

[0096] The embodiment discloses an image noise suppression system for laser speckle, comprising:

[0097] An image acquisition module is configured to acquire two laser images P1 and P2 at the same position.

[0098] An image decomposition module is configured to perform two-dimensional discrete wavelet transform on the two laser images P1 and P2 to obtain high-frequency images HH1, HL1 and LH1 of the laser image P1, a low-frequency image LL1 of the laser image P1, high-frequency images HH2, HL2 and LH2 of the laser image P2, and a low-frequency image LL2 of the laser image P2.

[0099] An image fusion module is configured to fuse the high-frequency images and the low-frequency images of the laser images P1 and P2 to obtain high-frequency fusion images HH', HL' and LH' and a low-frequency fusion image LL'.

[0100] A high-frequency fusion image filtering module is configured to use an optimized non-local mean filter algorithm to filter the high-frequency fusion images HH', HL' and LH' respectively to obtain filtered high-frequency fusion images HH'', HL'' and LH''.

[0101] An image restoration module is configured to perform wavelet inverse transform on the low-frequency fusion image LL' and the filtered high-frequency fusion images HH'', HL'' and LH'' to obtain a restored image P'.

[0102] a secondary filtering module, configured to filter the restored image P' using an optimized non-local mean filtering algorithm to obtain a final result image.

[0103] Embodiment 3

[0104] In this embodiment, the original laser image collected by the laser imaging system, the image with noise reduced by the non-local mean algorithm, and the image processed by the method of the present application are evaluated in terms of image quality, and the speckle suppression index (SSI) and entropy (EN) are used for comparison. SSI is used to evaluate the speckle suppression capability of the filtering algorithm, and the smaller the value, the stronger the suppression capability and the better the image quality. EN represents the amount of information carried by the image, and the larger the value, the more information retained by the image. The image processing results are shown in Table 1.

[0105] Table 1 shows the comparison of evaluation index values before and after noise reduction of the laser image

[0106]

[0107] As shown in Table 1, the method of the present application not only eliminates speckle noise, but also retains more information in the image compared to the traditional non-local mean filtering algorithm. Figures 5(a)-5(c) From the subjective visual point of view, the processing result of the method of the present application is better, and the speckle noise in the background can be eliminated. In order to observe more clearly, Figures 6(a)-6(c) The local magnification comparison results of the image before and after filtering.

[0108] The embodiments in the specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts of each embodiment can be referred to each other. For the system modules disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method part.

[0109] The above description of the disclosed embodiments enables a person skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to these embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for suppressing image noise caused by laser speckle, characterized in that, Includes the following steps: Step 1: Acquire two laser images P1 and P2 at the same location; Step 2: Perform two-dimensional discrete wavelet transform on the two laser images P1 and P2 to obtain the high-frequency images HH1, HL1, and LH1 of laser image P1, the low-frequency image LL1 of laser image P1, the high-frequency images HH2, HL2, and LH2 of laser image P2, and the low-frequency image LL2 of laser image P2. Step 3: Fuse the high-frequency and low-frequency images of laser images P1 and P2 to obtain high-frequency fused images HH', HL', LH' and low-frequency fused image LL'; In step 3, the specific method for fusing the high-frequency and low-frequency images of laser images P1 and P2 is as follows: For low-frequency images: the low-frequency image LL1 of laser image P1 and the low-frequency image LL2 of laser image P2 are fused using the averaging method. ; For high-frequency images: The weighted fusion method is used to fuse the high-frequency images HH1 of laser image P1 and HH2 of laser image P2, respectively; the high-frequency images HL1 of laser image P1 and HL2 of laser image P2 are fused; and the high-frequency images LH1 of laser image P1 and LH2 of laser image P2 are fused. The specific process of fusing high-frequency images using the weighted value fusion method is as follows: Calculate the weight values ​​of the two high-frequency images respectively. , ; Based on the weight values ​​of two high-frequency images , Solve for the fusion coefficient matrix of the two high-frequency images; The two high-frequency images are fused using the fusion coefficient matrix. The formula for calculating the weight value is as follows: ,in, This is the result after mean filtering of the high-frequency image. This is the result of median filtering on a high-frequency image; The fusion coefficient matrix is ​​expressed as: ； ； Step 4: Use the optimized nonlocal mean filtering algorithm to filter the high-frequency fused images HH', HL', and LH' respectively to obtain the filtered high-frequency fused images HH'', HL'', and LH''. In step 4, the optimized nonlocal mean filtering algorithm is based on the traditional nonlocal mean filtering algorithm, but with an improvement on the weight function. The improved weight function is derived from... express: in, n These are weight control parameters; Indicates the Euclidean distance between neighborhoods; Z ( x ) is the normalization coefficient. h For smoothing parameters; Step 5: By using inverse wavelet transform, the low-frequency fused image LL' and the filtered high-frequency fused images HH'', HL'', and LH'' are restored to obtain the restored image P'; Step 6: Use the optimized nonlocal mean filtering algorithm to filter the restored image P' to obtain the final result image.

2. The image noise suppression method for laser speckle according to claim 1, characterized in that, The Euclidean distance between the neighborhoods is expressed as: The normalization coefficient is expressed as: ； in, ds The radius of the neighborhood window; d= 2 ds+ 1; S t An image representing a hypothetical pixel difference; x 1, x 2 represents different pixels x coordinate.

3. An image noise suppression system for laser speckle, characterized in that, include: The image acquisition module is used to acquire two laser images P1 and P2 at the same location; The image decomposition module is used to perform two-dimensional discrete wavelet transform on the two laser images P1 and P2 to obtain the high-frequency images HH1, HL1, and LH1 of laser image P1, the low-frequency image LL1 of laser image P1, the high-frequency images HH2, HL2, and LH2 of laser image P2, and the low-frequency image LL2 of laser image P2. The image fusion module is used to fuse the high-frequency and low-frequency images of laser images P1 and P2 to obtain high-frequency fused images HH', HL', LH' and low-frequency fused image LL'. The specific method for fusing the high-frequency and low-frequency images of laser images P1 and P2 is as follows: For low-frequency images: the low-frequency image LL1 of laser image P1 and the low-frequency image LL2 of laser image P2 are fused using the averaging method. ; For high-frequency images: The weighted fusion method is used to fuse the high-frequency images HH1 of laser image P1 and HH2 of laser image P2, respectively; the high-frequency images HL1 of laser image P1 and HL2 of laser image P2 are fused; and the high-frequency images LH1 of laser image P1 and LH2 of laser image P2 are fused. The specific process of fusing high-frequency images using the weighted value fusion method is as follows: Calculate the weight values ​​of the two high-frequency images respectively. , ; Based on the weight values ​​of two high-frequency images , Solve for the fusion coefficient matrix of the two high-frequency images; The two high-frequency images are fused using the fusion coefficient matrix. The formula for calculating the weight value is as follows: ,in, This is the result after mean filtering of the high-frequency image. This is the result of median filtering on a high-frequency image; The fusion coefficient matrix is ​​expressed as: ； ； The high-frequency fusion image filtering module is used to filter the high-frequency fusion images HH', HL', and LH' respectively using an optimized nonlocal mean filtering algorithm to obtain the filtered high-frequency fusion images HH'', HL'', and LH''. The optimized nonlocal mean filtering algorithm is based on the traditional nonlocal mean filtering algorithm, with improvements made to the weight function. The improved weight function is derived from... express: in, n These are weight control parameters; Indicates the Euclidean distance between neighborhoods; Z ( x ) is the normalization coefficient. h For smoothing parameters; The image restoration module is used to restore the low-frequency fused image LL' and the filtered high-frequency fused images HH'', HL'', and LH'' through wavelet inverse transform, and obtain the restored image P'. The secondary filtering module is used to filter the reconstructed image P' using an optimized nonlocal mean filtering algorithm to obtain the final result image.

Images