Image denoising method and device based on statistical analysis

Abstract

The application provides a kind of based on statistical analysis image denoising method and device, it is related to image processing technical field, wherein method includes the following steps: laser phase unwrapping image is handled and analyzed, and training image data set is established;Image denoising model is constructed, and the image denoising model is trained using the training image data set;Actual laser image is acquired, and the image denoising model trained is used to denoising;The application establishes the noise model of laser imaging by theoretical analysis and experimental verification, and simulates the phase unwrapping image without noise, to construct corresponding data set, and the training image data set established is used to train image denoising model, subsequent image denoising model is used to actual phase unwrapping image denoising processing, can effectively realize the noise suppression of real phase unwrapping image, significantly improves the performance of laser imaging denoising.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, and more particularly to an image denoising method and device based on statistical analysis. BACKGROUND

[0002] Currently, laser imaging technology forms observable fringe patterns on the surface of an object through a laser projector, then captures these laser fringe images by an imaging device, and generates point clouds using reconstruction techniques. However, in this process, noise caused by coherence can introduce significant errors in the laser fringe images, phase wrapping images and phase unwrapping images, thereby reducing the clarity of the images. Ideally, the phase unwrapping image should be continuous in space and have a predictable trend, but the interference of noise can cause discontinuous jumps and abnormal phase changes, which in turn affect the final three-dimensional data. Therefore, the denoising processing of the phase unwrapping image is particularly important to reduce the impact of noise on data.

[0003] Researchers have proposed various filtering techniques to address the above problems, which can be broadly divided into three categories: spatial domain filtering, transform domain filtering and deep learning filtering. Spatial domain filtering is mainly based on the statistical properties of images and processes the entire image, including mean filtering, median filtering, Gaussian filtering, bilateral filtering and Kuan filtering, etc. Transform domain filtering methods involve domain transformation of images, such as Fourier transform and wavelet transform, as well as model-based denoising methods such as BM3D and WNNM, which can flexibly handle different noise levels. With the rapid development of deep learning technology, filtering models based on neural networks, such as DnCNN, FFDNet, CBDNet, RIDNet and PMRID, have become a research hotspot in this field. Although deep learning filtering algorithms are constantly emerging, their performance is to some extent dependent on high-quality training data, and most of their data sets come from image data sets such as Berkeley Segmentation Data Set. However, existing BSD data sets are not suitable for specific filtering tasks, and creating a data set suitable for a specific filtering task is both complex and time-consuming.

[0004] Therefore, how to provide an image denoising method based on statistical analysis to solve the above problems is a problem that those skilled in the art need to solve. SUMMARY

[0005] Therefore, the present application provides an image denoising method and device based on statistical analysis, which establishes a noise model of laser imaging through theoretical analysis and experimental verification, simulates a noise-free phase unwrapping image, constructs a corresponding data set, trains a denoising model through the data set, and subsequently uses the denoising model for denoising, which can effectively realize noise suppression of real phase unwrapping images and significantly improve the performance of laser imaging denoising.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] An image denoising method based on statistical analysis comprises the following steps:

[0008] Acquire laser images and phase unwrapped images generated in laser imaging for processing and analysis, and construct a training image dataset;

[0009] Constructing an image denoising model, and training the image denoising model using the training image dataset;

[0010] An actual phase unwrapped image is obtained, and denoised using the trained image denoising model.

[0011] Preferably, the specific process of constructing the training image dataset includes:

[0012] Processing the laser image to obtain a corresponding echo signal;

[0013] Processing the echo signal to obtain a corresponding noise model;

[0014] The phase unwrapped image is analyzed in combination with the noise model, and a training image dataset is constructed according to the analysis results.

[0015] Preferably, the specific process of processing the echo signal includes:

[0016] Analyzing the echo signal to obtain a probability density function corresponding to the amplitude of the echo signal;

[0017] Determining a probability density function of an additive noise signal in the echo signal according to the probability density function;

[0018] The phase unwrapping image is fitted, and the fitting result, the laser image and the probability density function of the additive noise signal are logarithmically transformed and point-by-point subtracted to determine a noise model.

[0019] Preferably, the specific process of processing the echo signal further includes:

[0020] The noise model is verified.

[0021] Preferably, the specific process of determining the probability density function of the additive noise signal contained in the echo signal includes:

[0022] Determining a probability density function of pixel intensities included in the phase unwrapped image according to the probability density function of the amplitude;

[0023] A probability density function of the original noise is determined, and a probability density function of the additive noise signal is determined according to an additive noise model.

[0024] Preferably, the image denoising model is a DnCNN model, comprising an input layer, a hidden layer and an output layer connected in sequence.

[0025] The application also provides an image denoising device based on statistical analysis, comprising:

[0026] A data set construction module is configured to process and analyze laser images and phase unwrapping images generated in laser imaging, and construct a training image data set;

[0027] A model construction module is configured to construct an image denoising model, and train the image denoising model using the training image data set;

[0028] A processing module is configured to acquire an actual laser image, and denoise the actual laser image using the trained image denoising model.

[0029] Compared with the prior art, the application discloses an image denoising method and device based on statistical analysis, which can effectively filter noise by efficiently suppressing specific Gumbel noise in an image. The algorithm is optimized for phase unwrapping images, and the performance of laser imaging denoising is significantly improved through accurate statistical analysis and deep learning technology.

[0030] There are currently various deep learning-based image filtering models, such as DnCNN model, FFDNet model, CBDNet model, RIDNet model and PMRID network model. Although these deep learning filtering algorithms are constantly emerging, their performance largely depends on high-quality training data, and these data sets are mostly derived from Berkeley segmentation data sets. Creating a data set specific to the filtering task is usually complex and time-consuming, increasing the difficulty of implementation.

[0031] The application reveals that the noise model in laser imaging follows Gumbel distribution through in-depth theoretical derivation, and develops a specific phase unwrapping image data set, which, combined with a classic DnCNN network model, achieves the purpose of three-dimensional point cloud filtering through phase unwrapping image denoising. BRIEF DESCRIPTION OF DRAWINGS

[0032] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments or the prior art description will be briefly introduced below. Obviously, the drawings in the following description are only embodiments of the present application, and those skilled in the art can obtain other drawings according to the provided drawings without creative labor.

[0033] Figure 1 A flow chart of the image denoising method based on statistical analysis provided by the present application is shown in the figure;

[0034] Figure 2 A three-dimensional laser imaging principle diagram provided by the embodiment of the present application is shown in the figure;

[0035] Figure 3 A target noise distribution diagram provided by the embodiment of the present application is shown in the figure, where a represents the three-dimensional display of the original data of the plane and the fitting plane, and b represents the noise intensity distribution of the plane;

[0036] Figure 4 A DnCNN network structure diagram provided by the embodiment of the present application is shown in the figure;

[0037] Figure 5 A structure principle diagram of the image denoising device based on statistical analysis provided by the present application is shown in the figure. DETAILED DESCRIPTION

[0038] 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 a 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 labor fall within the scope of protection of the present application.

[0039] Referring to Figure 1 The embodiment of the present application discloses an image denoising method based on statistical analysis, which comprises the following steps:

[0040] The laser image and the phase unwrapping image generated in laser imaging are acquired for processing and analysis, and a training image data set is constructed;

[0041] An image denoising model is constructed, and the image denoising model is trained by using the training image data set;

[0042] An actual phase unwrapping image is acquired, and the image denoising model after training is used for denoising.

[0043] In a specific embodiment, the specific process of constructing the training image data set comprises:

[0044] The laser image is processed to obtain the corresponding echo signal;

[0045] The echo signal is processed to obtain the corresponding noise model;

[0046] The phase unwrapping image is analyzed in combination with the noise model, and the training image data set is constructed according to the analysis result.

[0047] In one specific embodiment, the specific process of processing the echo signal comprises:

[0048] analyzing the echo signal to obtain a probability density function corresponding to the amplitude of the echo signal;

[0049] determining a probability density function of the additive noise signal contained in the echo signal according to the probability density function;

[0050] fitting the phase unwrapping image, and performing logarithmic transformation and point-by-point subtraction on the fitting result, the laser image, and the probability density function of the additive noise signal to determine a noise model.

[0051] In one specific embodiment, the specific process of processing the echo signal further comprises:

[0052] verifying the noise model.

[0053] In one specific embodiment, the specific process of determining the probability density function of the additive noise signal contained in the echo signal comprises:

[0054] determining a probability density function of pixel intensity included in the phase unwrapping image according to the probability density function of the amplitude;

[0055] determining a probability density function of the original noise, and determining the probability density function containing the additive noise signal according to the additive noise model.

[0056] Specifically, Figure 2 A schematic diagram for theoretical analysis of three-dimensional laser imaging is shown in FIG. 1. When light pulses emitted by a light source propagate in multiple directions, they encounter irregular scatterers and uneven interfaces, generating scattered light waves different from the original propagation direction. Part of the scattered light waves will return in the form of backscattering and be captured by the receiver. Since the returned scattered light waves come from the same light source, they follow specific statistical rules. The echo generated by each scatterer can be represented as a complex number The received total echo signal is considered as the superposition of multiple echoes, and the specific expression is:

[0057]

[0058] In the formula, denotes the echo of the kth scatterer, A k and θ k are the amplitude and phase, respectively, and n is the total number of scatterers. The total echo signal can be represented by a complex number Z=A r +jA i =Acosθ+jAsinθ. When n is large, the real part A r and the imaginary part A iapproximately follow a Gaussian distribution with mean 0 and standard deviation σ. The amplitude A of the echo signal is calculated by equation (2):

[0059]

[0060] where A r 2 and A i 2 will follow a chi-square distribution with one degree of freedom. Therefore, the probability density function of the echo signal amplitude will follow a Rayleigh distribution, which is given by:

[0061]

[0062] where σ is the standard deviation. The pixel intensity I is calculated by I = A 2 and the probability density function of I is given by:

[0063]

[0064] The model of the large amount of multiplicative noise caused by the coherence between the lasers is:

[0065] G(x,y) = F(x,y)N(x,y) (5)

[0066] where G(x,y) is the noisy laser image, F(x,y) is the noise-free expected image, and N(x,y) is the noise component. A log transformation is used to convert the multiplicative noise into additive noise form to simplify the noise model. The model is given by:

[0067] G'(x,y) = F'(x,y) + N'(x,y) (6)

[0068] where G'(x,y), F'(x,y), and N'(x,y) are the log-transformed laser noise image, the log-transformed noise-free expected image, and the log-transformed noise component, respectively. The form of N' is given by:

[0069] N' = aln(N) + b (7)

[0070] where a and b are coefficients. The probability density function f N′ (n') of N' is obtained by transforming the probability density function of the original noise N:

[0071]

[0072] In the imaging process, the noise usually comes from the superposition of a large amount of scattered light waves generated by the same light source, so they follow the same statistical law as the echo signal. The probability density function of the original noise is given by:

[0073]

[0074] The specific expression of the probability density function of additive noise is:

[0075]

[0076] The original noise distribution is logarithmically transformed and its probability density function is obtained, which conforms to the Gumbel model equation.

[0077] In order to verify the noise model, a flat plate is selected as the target object, which provides ideal test conditions for the experiment due to its simple geometric characteristics and controllable surface reflection properties. This design can accurately evaluate and verify the statistical characteristics of image noise, providing a solid foundation for the development and optimization of subsequent noise suppression algorithms.

[0078] First, it can be reasonably assumed that the phase unwrapping image mainly contains noise components and useful phase change signals. The least squares optimization algorithm is used to fit the image data with a polynomial equation, and the fitted plane is considered as the noise-free useful signal:

[0079] Z = a0 + a1X + a2Y + a3X 2 + a4XY + a5Y 2 + a6X 3 + a7X 2 Y + a8XY 2 + a9Y 3 (11)

[0080] In the formula, Z is the dependent variable, X and Y are independent variables, and a k (k = 1, 2,..., 9) are model parameters. As Figure 3 (a) is a three-dimensional display of the original data and the fitted data. The original data and the fitted data are logarithmically transformed and subtracted point by point to extract the noise component after logarithmic transformation, and the specific expression is:

[0081] log[N(x, y)] = log[G(x, y)] - log[F(x, y)] (12)

[0082] After cleaning the data, the Maximum Likelihood Estimation (MLE) algorithm is used to statistically analyze the noise data and fit the Gaussian distribution and Gumbel distribution. As shown in Figure 3 (b), the Gaussian distribution shows a classic symmetrical bell-shaped curve, while the Gumbel distribution shows obvious skewness, with a slower decay rate in the tail. This verifies the consistency between the laser noise data after logarithmic transformation and the Gumbel distribution.

[0083] Subsequently, in constructing the training image dataset, the training image dataset contains 400 high-depth (16-bit) phase unwrapping images, and these high-precision data help to improve the sensitivity of the algorithm to image noise, thereby improving the accuracy of the reconstruction. The image size of the training image dataset is 180x180 pixels. Based on the analysis of the actual laser phase unwrapping image, the pixel value increases from left to right in steps of 30-50, and the step size of the simulated phase unwrapping image is set to Step∈(30, 50) to simulate the increasing trend of the pixel. At the same time, considering the depth difference of the target object, there will be a pixel jump Jump in the image, and its range is set to Jump∈(-30000, 30000) to reflect the pixel mutation caused by the depth difference of the target. In addition, due to the non-parallelism of the actual target relative to the reference surface, a random number is added to the original phase tilt, and its range is number∈(-100, 100) to simulate the uncertainty of the target phase change. The invention sets the noise scale parameter range σ∈[0.001, 0.01] to cover various noise levels that may be encountered in actual imaging. This model is designed to suppress Gumbel noise in phase unwrapping images to achieve more accurate data reconstruction in three-dimensional imaging.

[0084] In a specific embodiment, the image denoising model is a DnCNN model, which includes an input layer, a hidden layer and an output layer connected in sequence.

[0085] Specifically, referring to Figure 4 As shown in the figure, the embodiment of the present application adopts a deep convolutional neural network (Denoising Convolutional Neural Network, DnCNN) to effectively suppress the noise in the phase unwrapping image. The network depth is set to 17, aiming to obtain rich spatial features. The network structure is composed of three main parts: the input layer uses a convolution layer and a ReLU activation function to preliminarily extract image features, and configures 64 convolution kernels to generate 64 feature channels. The hidden layer is composed of multiple convolution layers, batch normalization layers and ReLU activation functions, which further extract and fuse image features, and all layers use 3x3 convolution kernels. The output layer linearly combines 64 feature channels through a convolution layer to reconstruct a filtered image. Each convolution layer is followed by a ReLU activation function to introduce nonlinearity to enhance the model's ability to learn complex features. The batch normalization processing in the hidden layer effectively reduces the internal covariate shift and speeds up the convergence speed of the model.

[0086] The goal of network training is to learn a residual mapping R(y) (y represents a noisy image) so that the residual image R(y) predicted by the network is close to the true residual image v (v = yt, t is the true image), and finally obtain the ideal image x through x = yR(y). During the training process, the mean square error (MSE) is used as the loss function to optimize the residual mapping. The specific expression is:

[0087]

[0088] Where M is the number of training samples; v i is the true residual of the i-th phase unwrapped image; R i (y i ) is the residual result of network prediction.

[0089] See also Figure 5 As shown, an embodiment of the present invention provides an apparatus for using the image denoising method based on statistical analysis described in any one of the above embodiments, including:

[0090] A data set construction module is used to process and analyze the laser images and phase unwrapped images generated in laser imaging, and to construct a training image data set;

[0091] A model building module is used to build an image denoising model and train the image denoising model using a training image dataset;

[0092] The processing module is used to obtain the actual laser image and perform denoising using the trained image denoising model.

[0093] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the devices 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 description.

[0094] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An image denoising method based on statistical analysis, characterized in that: The following steps are involved: The laser image and the corresponding phase unwrapped image generated in laser imaging are acquired for processing and analysis, and a training image dataset is constructed. The specific process includes: Processing the laser image to obtain a corresponding echo signal; Processing the echo signal to obtain a corresponding noise model; Analyzing the phase unwrapped image in combination with the noise model, and constructing a training image dataset based on the analysis results; The specific process of processing the echo signal includes: Analyzing the echo signal to obtain a probability density function corresponding to the amplitude of the echo signal; Determining a probability density function of an additive noise signal in the echo signal according to the probability density function; Fitting the phase unwrapped image, performing logarithmic transformation and point-by-point subtraction on the probability density functions of the fitting result, the laser image, and the additive noise signal to determine a noise model; The specific process of determining the probability density function of the additive noise signal included in the echo signal includes: Determining a probability density function of pixel intensities included in the phase unwrapped image according to the probability density function of the amplitude; determining a probability density function of the original noise and determining a probability density function of the signal including the additive noise based on the additive noise model; Constructing an image denoising model, and training the image denoising model using the training image dataset; An actual phase unwrapped image is obtained, and denoised using the trained image denoising model.

2. The image denoising method based on statistical analysis according to claim 1, characterized in that: The specific process of processing the echo signal further includes: The noise model is verified.

3. The image denoising method based on statistical analysis according to claim 1, characterized in that: The image denoising model is a DnCNN model, which includes an input layer, a hidden layer and an output layer connected in sequence.

4. A device using the image denoising method based on statistical analysis according to any one of claims 1 to 3, characterized in that: include: The dataset construction module is used to process and analyze the laser images and phase unwrapped images generated by laser imaging and construct a training image dataset. The specific process includes: Processing the laser image to obtain a corresponding echo signal; Processing the echo signal to obtain a corresponding noise model; Analyzing the phase unwrapped image in combination with the noise model, and constructing a training image dataset based on the analysis results; The specific process of processing the echo signal includes: Analyzing the echo signal to obtain a probability density function corresponding to the amplitude of the echo signal; Determining a probability density function of an additive noise signal in the echo signal according to the probability density function; Fitting the phase unwrapped image, performing logarithmic transformation and point-by-point subtraction on the probability density functions of the fitting result, the laser image, and the additive noise signal to determine a noise model; The specific process of determining the probability density function of the additive noise signal included in the echo signal includes: Determining a probability density function of pixel intensities included in the phase unwrapped image according to the probability density function of the amplitude; determining a probability density function of the original noise and determining a probability density function of the signal including the additive noise based on the additive noise model; A model building module, configured to build an image denoising model and train the image denoising model using the training image dataset; The processing module is used to obtain an actual laser image and perform denoising using the trained image denoising model.

Images