RAW format image denoising method, system and device and storage medium

Abstract

The invention relates to a RAW format image denoising method and device, equipment and a storage medium. The method comprises the following steps: acquiring an RAW format image, and extracting linear features of the RAW format image; and determining a parallelogram according to the linear features, and selecting a central region of the parallelogram as a reference image block to carry out Gaussian noise estimation to obtain a histogram and an estimation parameter of the reference image block. And based on a block matching three-dimensional filtering algorithm, taking the estimation parameters as hyper-parameters, constructing a denoising model, inputting the RAW format image into the denoising model for filtering processing, and outputting a denoised target image. By adopting the method, the image signal denoising precision and real-time performance can be improved.

Description

Technical Field

[0001] This invention relates to the field of image signal processing technology, and in particular to a method, apparatus, device, and storage medium for denoising RAW format images. Background Technology

[0002] In the image processing of an ISP (Image Signal Processor), noise is a key factor leading to a decline in image quality and severely affecting the quality of data perception. Noise arises from the inherent characteristics of the image sensor, the operating environment, and interference during signal transmission and storage. For example, CMOS and CCD sensors introduce various types of noise during image acquisition, such as thermal noise and photon noise; these can be categorized by pattern (fixed-pattern noise, random noise, etc.) and by probability distribution (Gaussian noise, Rayleigh noise, etc.).

[0003] Current typical image denoising methods encompass both traditional and deep learning approaches. Traditional denoising methods include spatial domain denoising such as mean filtering, median filtering, and Gaussian filtering, while transform domain denoising includes low-pass filtering and wavelet filtering. Other methods include non-local mean filtering based on image self-similarity and BM3D (Block-Matching and 3D filtering). Deep learning methods utilize models such as multilayer perceptrons, autoencoders, and convolutional neural networks to achieve denoising. The noise in RAW images output from ISP sensors can be described by a Gaussian-Poisson model, which can be used to optimize denoising algorithms.

[0004] However, existing technologies all have certain problems. In traditional spatial domain denoising methods, mean filtering, while reducing noise amplitude, expands the area contaminated by noise; Gaussian filtering, because its weights depend only on Euclidean distance, easily loses image edge information; median filtering, while preserving edges well, leads to the loss of small targets and image discontinuities. Transform domain denoising methods, if the filtering domain, filter, and parameters are not properly selected, can easily cause loss of image details or textures. Deep learning denoising methods, while highly effective, have high requirements for computing power and storage resources, making it difficult to meet real-time requirements and limiting their application scenarios. Summary of the Invention

[0005] Therefore, it is necessary to provide a method, apparatus, device, and storage medium for denoising RAW format images that can improve the denoising accuracy and real-time performance of image signals, addressing the aforementioned technical problems.

[0006] A method for denoising RAW format images, the method comprising: Obtain a RAW format image and extract the straight line features from it.

[0007] Based on the characteristics of straight lines, a parallelogram is determined. The central region of the parallelogram is selected as a reference image block for Gaussian noise estimation, and the histogram and estimated parameters of the reference image block are obtained.

[0008] Based on the block matching 3D filtering algorithm, the estimated parameters are used as hyperparameters to construct a denoising model. The RAW format image is input into the denoising model for filtering and the denoised target image is output.

[0009] In one embodiment, the method further includes: acquiring a RAW format image corresponding to a standard color chart using an image sensor; mapping randomly selected pixels from the RAW format image data to a polar coordinate plane and then accumulating them to obtain an accumulation plane; obtaining the line corresponding to the local maxima that meet the threshold requirements based on the accumulation plane; and searching for line features in the Cartesian coordinate system corresponding to the two endpoints of the line.

[0010] In one embodiment, the method further includes: selecting lines whose segment length is equal to the minimum threshold length from the line features to form a parallelogram, and determining the central region of the parallelogram based on the size of a preset standard pixel block.

[0011]

[0012]

[0013] in, The x-axis coordinates of the center point of the central region. The vertical coordinate of the center point of the central region is y = y. Let be the coordinates of the top left vertex of the parallelogram. Let be the coordinates of the top right vertex of the parallelogram. Let be the coordinates of the lower left vertex of the parallelogram. Let be the coordinates of the lower right vertex of the parallelogram. , , , These are the coordinates of the four vertices of the central region: top left, top right, bottom left, and bottom right. The pixel block size.

[0014] In one embodiment, the method further includes: selecting the central region of a parallelogram as a reference image block for Gaussian noise estimation, and calculating the mean of the central region serving as the reference image block.

[0015] in, The mean of the reference image patch, The total number of pixels in the reference image patch. To select the number of sub-bins, for The order, The reference image patch. And the variance:

[0016] in, The variance of the reference image patch is given. The histogram of the reference image patch is obtained. The Gaussian distribution of the central region is estimated based on the mean and variance, and the probability density function is obtained.

[0017] in, Let be the probability density function. The input pixel values ​​are used. The histogram is analyzed and compared with the probability density function. If the comparison results are similar, the mean and variance are used as estimation parameters for the Gaussian noise corresponding to the reference image patch.

[0018] In one embodiment, the method further includes: removing the endpoints of the line segments that make up the line features from the RAW format image, and recording the coordinates of each endpoint.

[0019] In one embodiment, the method further includes: analyzing the alignment histogram and probability density function, arranging the alignment results in order, and taking the mean and variance of the alignment results that are sorted within the standard interval as the estimation parameters of the Gaussian noise corresponding to the reference image block.

[0020] In one embodiment, the method further includes: constructing a denoising model based on a block matching 3D filtering algorithm, using the estimated parameters as hyperparameters; inputting a RAW format image into the denoising model to complete image block segmentation and similar block search, 3D transformation and filtering, and inverse transformation and aggregation filtering processes; enhancing the image through histogram equalization; and outputting a denoising objective function.

[0021] A noise reduction apparatus for RAW format images, the apparatus comprising: The geometric feature acquisition module is used to acquire RAW format images and extract straight line features from them.

[0022] The Gaussian noise estimation module is used to determine the parallelogram based on the straight line characteristics, select the central region of the parallelogram as the reference image block for Gaussian noise estimation, and obtain the histogram and estimated parameters of the reference image block.

[0023] The denoising module is used to construct a denoising model based on the block matching 3D filtering algorithm, using the estimated parameters as hyperparameters. The RAW format image is input into the denoising model for filtering and the denoised target image is output.

[0024] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program performing the following steps: Obtain a RAW format image and extract the straight line features from it.

[0025] Based on the characteristics of straight lines, a parallelogram is determined. The central region of the parallelogram is selected as a reference image block for Gaussian noise estimation, and the histogram and estimated parameters of the reference image block are obtained.

[0026] Based on the block matching 3D filtering algorithm, the estimated parameters are used as hyperparameters to construct a denoising model. The RAW format image is input into the denoising model for filtering and the denoised target image is output.

[0027] A computer-readable storage medium having a computer program stored thereon, the computer program performing the following steps when executed by a processor: Obtain a RAW format image and extract the straight line features from it.

[0028] Based on the characteristics of straight lines, a parallelogram is determined. The central region of the parallelogram is selected as a reference image block for Gaussian noise estimation, and the histogram and estimated parameters of the reference image block are obtained.

[0029] Based on the block matching 3D filtering algorithm, the estimated parameters are used as hyperparameters to construct a denoising model. The RAW format image is input into the denoising model for filtering and the denoised target image is output.

[0030] The aforementioned method, apparatus, device, and storage medium for denoising RAW format images first extracts straight line features from the RAW format image. Compared to the traditional Hough transform, this algorithm reduces computational redundancy through probability sampling, significantly improving straight line detection efficiency and laying a real-time foundation for subsequent processing. Simultaneously, accurate extraction of straight line features helps determine the parallelogram, avoiding denoising deviations caused by false edge detection and indirectly optimizing denoising accuracy. In the noise estimation stage, the central region of the parallelogram is selected as a reference image block. This region typically contains the core structural information of the image and is less affected by edge interference, effectively avoiding the problems of median filtering losing small targets and mean filtering expanding the noise area. Gaussian noise estimation is performed based on this region to obtain accurate histograms and estimated parameters, providing a reliable hyperparameter basis for the block matching 3D filtering (BM3D) algorithm. Compared to transform domain denoising that relies on empirical parameter selection, this method significantly reduces the risk of detail loss due to inappropriate parameters. In the denoising stage, the BM3D algorithm is constructed using estimated parameters to drive the denoising model. This algorithm utilizes image self-similarity for 3D collaborative filtering and fully considers the correlation of image content through a block matching mechanism, effectively solving the problem of edge loss caused by Gaussian filtering ignoring image semantics. Furthermore, the BM3D algorithm has controllable complexity and, compared to deep learning methods, can achieve efficient denoising without relying on high-performance hardware, meeting real-time requirements while ensuring denoising accuracy. The entire scheme forms a complete "feature extraction-noise estimation-efficient denoising" chain through asymptotic probabilistic Hough transform for rapid location of key structures, parallelogram region screening for optimized noise estimation, and precise hyperparameter-driven BM3D filtering. This improves denoising accuracy and real-time performance in RAW format image denoising. Attached Figure Description

[0031] Figure 1 This is a flowchart illustrating a method for denoising a RAW format image in one embodiment. Figure 2 This is a schematic diagram of a BM3D image denoising process for RAW format ColorChecker images in an ISP process according to one embodiment. Figure 3 This is an image of 24 colored polygons in one embodiment; Figure 4 This is an image of 24 colored polygons before denoising in one embodiment; Figure 5 This is an image of 24 color polygons after denoising in one embodiment; Figure 6 This is a structural block diagram of a noise reduction device for a RAW format image in one embodiment; Figure 7 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0032] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0033] In one embodiment, such as Figure 1 As shown, a method for denoising RAW format images is provided, including the following steps: Step 102: Obtain the RAW format image and extract the straight line features from the RAW format image.

[0034] Specifically, existing image sensors (such as CMOS) can be used to acquire RAW format data of standard color charts (such as ColorChecker images). After acquiring the RAW format image data, pseudo-color processing can be performed on the image data to highlight information such as lines and edges in the image.

[0035] Furthermore, based on the RAW format data of the ColorChecker image containing several straight lines, N pixels are randomly selected from the set of edge points in the image data using asymptotic probability Hough transform. These pixels are then mapped to a cumulative plane of polar coordinates. Specifically, for a non-zero point in the image plane... ,make From 0 to The changes are calculated using the following formulas. value:

[0036] A series of pixels can be obtained, and these pixels form a certain point. The corresponding curve in the plane is mapped to After applying the polar coordinates to the image plane, summing all the results will display the straight line in the image plane as... Local maxima in a plane; Furthermore, when filtering local maxima that meet the threshold requirements in the cumulative plane, non-maximum suppression (NMS) is used to filter peak points, and a threshold (such as vote count > 20) is set to filter noise, several corresponding straight lines are obtained.

[0037] Furthermore, the endpoints of the line are detected and transformed into Cartesian coordinates, converting the polar coordinates... Converting to the Cartesian equation y=mx+b, the endpoints are located by calculating the intersections of the line with the image boundary and filtering out invalid intersections (outside the image range). The RANSAC algorithm can be used to optimize the endpoint positions, improve robustness, and fit the line segment.

[0038] Step 104: Determine the parallelogram based on the straight line characteristics, select the central region of the parallelogram as the reference image block for Gaussian noise estimation, and obtain the histogram and estimated parameters of the reference image block.

[0039] Specifically, by repeatedly performing step 102, several straight lines and their endpoints are obtained, ultimately leading to the coordinates of several colored parallelograms. The central region of each parallelogram is set as a 50*50 pixel block. The coordinates of the four vertices of the parallelogram—top left, top right, bottom left, and bottom right—are set as follows: The center point of the central area of The formula for calculating the axis coordinates is:

[0040] The formula for calculating the axis coordinates is:

[0041] The coordinates of the four vertices of the central region can then be calculated as follows:

[0042] Furthermore, for a given parallelogram, its center coordinates are calculated, and a fixed-size image patch with the center as the reference is extracted as a reference image patch. For the extracted reference image patch, the mean and variance parameters of Gaussian noise can be calculated using the maximum likelihood estimation (MLE) method, and its histogram can be generated.

[0043] Furthermore, based on the obtained mean and variance parameters, the Gaussian distribution of the image is estimated, and its probability density function can be expressed as: , in The input pixel values ​​are used, and the resulting PDF is compared with the histogram of the calculated reference image block. If the two are close, the calculated mean and variance can be used as the estimated parameters of Gaussian noise. This step and parameter calculation step are repeated to obtain multiple sets of estimated parameters of Gaussian noise. The estimated noise results are arranged from smallest to largest, and the mean of the middle four results is calculated as the Gaussian noise parameter.

[0044] Step 106: Based on the block matching 3D filtering algorithm, a denoising model is constructed using the estimated parameters as hyperparameters. The RAW format image is input into the denoising model for filtering and the denoised target image is output.

[0045] Specifically, the Gaussian noise parameter is used as a hyperparameter of BM3D and input together with the original noisy image into the BM3D module. Finally, after filtering, BM3D obtains the denoised image.

[0046] In the aforementioned denoising method for RAW format images, firstly, for RAW format images, asymptotic probabilistic Hough transform is used to extract line features. Compared to the traditional Hough transform, this algorithm reduces computational redundancy through probability sampling, significantly improving line detection efficiency and laying a real-time foundation for subsequent processing. Simultaneously, accurate extraction of line features helps determine parallelograms, avoiding denoising deviations caused by edge false detections and indirectly optimizing denoising accuracy. In the noise estimation stage, the central region of the parallelogram is selected as a reference image block. This region typically contains the core structural information of the image and is less affected by edge interference, effectively avoiding the problems of median filtering losing small targets and mean filtering expanding the noise area. Gaussian noise estimation is performed based on this region to obtain accurate histograms and estimated parameters, providing a reliable hyperparameter basis for the BM3D algorithm. Compared to transform domain denoising that relies on empirical parameter selection, this method significantly reduces the risk of detail loss due to inappropriate parameters. In the denoising stage, the BM3D algorithm is constructed using estimated parameters to drive the denoising model. This algorithm utilizes image self-similarity for 3D collaborative filtering and fully considers the correlation of image content through a block matching mechanism, effectively solving the problem of edge loss caused by Gaussian filtering ignoring image semantics. Furthermore, the BM3D algorithm has controllable complexity and, compared to deep learning methods, can achieve efficient denoising without relying on high-performance hardware, meeting real-time requirements while ensuring denoising accuracy. The entire scheme forms a complete "feature extraction-noise estimation-efficient denoising" chain through asymptotic probabilistic Hough transform for rapid location of key structures, parallelogram region screening for optimized noise estimation, and precise hyperparameter-driven BM3D filtering. This improves denoising accuracy and real-time performance in RAW format image denoising.

[0047] In one embodiment, an image sensor is used to acquire a RAW format image corresponding to a standard color chart. Randomly selected pixels from the RAW format image data are mapped to a polar coordinate plane and then accumulated to obtain an accumulation plane. Based on the accumulation plane, the lines corresponding to the local maxima that meet the threshold requirements are obtained, and the linear features of the two endpoints of the lines are searched in the Cartesian coordinate system.

[0048] In one embodiment, a parallelogram is formed by selecting straight lines whose segment length is equal to the minimum threshold length from the straight line features, and the central region of the parallelogram is determined according to the size of a preset standard pixel block.

[0049]

[0050]

[0051] in, The x-axis coordinates of the center point of the central region. The vertical coordinate of the center point of the central region is y = y. Let be the coordinates of the top left vertex of the parallelogram. Let be the coordinates of the top right vertex of the parallelogram. Let be the coordinates of the lower left vertex of the parallelogram. Let be the coordinates of the lower right vertex of the parallelogram. , , , These are the coordinates of the four vertices of the central region: top left, top right, bottom left, and bottom right. The pixel block size.

[0052] It's worth noting that the asymptotic probability voting mechanism only votes on a subset of edge points (e.g., 2000 / 10000 points), improving computational efficiency by more than 5 times. Threshold filtering (votes > 20) effectively filters noise, avoiding the global computational redundancy of traditional Hough transform. Considering the geometric characteristics of the standard color chart (parallelogram), edge lines are extracted first, providing accurate region boundaries for subsequent noise estimation. Experiments show that the color chart edge detection accuracy reaches 98.7%, significantly higher than the traditional Canny+Hough method (approximately 85%), while reducing memory usage by 40%, making it suitable for embedded device deployments (such as smartphone ISPs).

[0053] In one embodiment, the central region of a parallelogram is selected as a reference image block for Gaussian noise estimation, and the mean of the central region serving as the reference image block is calculated:

[0054] in, The mean of the reference image patch, The total number of pixels in the reference image patch. To select the number of sub-bins, for The order, The reference image patch. And the variance:

[0055] in, The variance of the reference image patch is given. The histogram of the reference image patch is obtained. The Gaussian distribution of the central region is estimated based on the mean and variance, and the probability density function is obtained.

[0056] in, Let be the probability density function. The input pixel values ​​are used. The histogram is analyzed and compared with the probability density function. If the comparison results are similar, the mean and variance are used as estimation parameters for the Gaussian noise corresponding to the reference image patch.

[0057] In one embodiment, the endpoints that make up the line segment in the line feature are removed from the RAW format image, and the coordinates of each endpoint are recorded.

[0058] In one embodiment, the alignment histogram and probability density function are analyzed and compared. After the alignment results are arranged in order, the mean and variance of the alignment results that are sorted within the standard interval are taken as the estimation parameters of the Gaussian noise corresponding to the reference image block.

[0059] In one embodiment, a denoising model is constructed based on a block matching 3D filtering algorithm, using the estimated parameters as hyperparameters. The RAW format image is input into the denoising model to complete image block segmentation and similar block search, 3D transformation and filtering, and inverse transformation and aggregation filtering. The image is then enhanced by histogram equalization, and the denoising objective function is output.

[0060] In one embodiment, such as Figure 2 As shown, this paper presents a BM3D (Block-Matching and 3D filtering) image denoising workflow for RAW format ColorChecker images in an ISP process. The specific content is as follows: Step 1: Acquire some RAW format data of ColorChecker images using the existing CMOS sensor; Step 2: Use the asymptotic probability Hough transform to detect lines and obtain lines from the ColorChecker; Step 3: Repeat step 2 continuously to obtain 96 straight lines and their endpoints, and finally obtain the coordinates of 24 colored parallelograms; Step 4: Select the central regions of each of the 24 parallelograms as reference pixel blocks for estimating Gaussian noise, and then perform noise estimation; Step 5: Arrange the 24 estimated noise results in ascending order, and calculate the mean of the four middle results as the Gaussian noise parameter; Step 6: Input the Gaussian noise parameter as a hyperparameter of BM3D along with the original noisy image into the BM3D module. Finally, BM3D will filter the image to obtain the denoised image.

[0061] It is worth noting that, such as Figure 3 The image shown is of the 24 colored polygons detected in step 3. Figure 4 The image before denoising using this method. Figure 5 The image after denoising using this method It should be understood that, although Figures 1-2 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figures 1-2 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0062] In one embodiment, such as Figure 6 As shown, a denoising device for RAW format images is provided, including: a geometric feature acquisition module 602, a Gaussian noise estimation module 604, and a denoising module 606, wherein: The geometric feature acquisition module 602 is used to acquire RAW format images and extract straight line features from the RAW format images.

[0063] The Gaussian noise estimation module 604 is used to determine the parallelogram based on the straight line characteristics, select the central region of the parallelogram as a reference image block for Gaussian noise estimation, and obtain the histogram and estimation parameters of the reference image block.

[0064] The denoising module 606 is used to construct a denoising model based on the block matching 3D filtering algorithm, using the estimated parameters as hyperparameters. The RAW format image is input into the denoising model for filtering and the denoised target image is output.

[0065] For specific limitations regarding a denoising device for RAW format images, please refer to the limitations of a denoising method for RAW format images mentioned above, which will not be repeated here. Each module in the aforementioned denoising device for RAW format images can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0066] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 7As shown, the computer device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with external terminals via a network connection. When executed by the processor, the computer program implements a noise reduction method for RAW format images. The display screen can be a liquid crystal display (LCD) or an e-ink display. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse.

[0067] Those skilled in the art will understand that Figures 6-7 The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the computer device to which the present invention is applied. A specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0068] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to perform the following steps: Obtain a RAW format image and extract the straight line features from it.

[0069] Based on the characteristics of straight lines, a parallelogram is determined. The central region of the parallelogram is selected as a reference image block for Gaussian noise estimation, and the histogram and estimated parameters of the reference image block are obtained.

[0070] Based on the block matching 3D filtering algorithm, the estimated parameters are used as hyperparameters to construct a denoising model. The RAW format image is input into the denoising model for filtering and the denoised target image is output.

[0071] In one embodiment, when the processor executes the computer program, it further performs the following steps: selecting the central region of a parallelogram as a reference image block for Gaussian noise estimation, and calculating the mean of the central region serving as the reference image block:

[0072] in, The mean of the reference image patch, The total number of pixels in the reference image patch. To select the number of sub-bins, for The order, The reference image patch. And the variance:

[0073] in, The variance of the reference image patch is given. The histogram of the reference image patch is obtained. The Gaussian distribution of the central region is estimated based on the mean and variance, and the probability density function is obtained.

[0074] in, Let be the probability density function. The input pixel values ​​are used. The histogram is analyzed and compared with the probability density function. If the comparison results are similar, the mean and variance are used as estimation parameters for the Gaussian noise corresponding to the reference image patch.

[0075] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor: Obtain a RAW format image and extract the straight line features from it.

[0076] Based on the characteristics of straight lines, a parallelogram is determined. The central region of the parallelogram is selected as a reference image block for Gaussian noise estimation, and the histogram and estimated parameters of the reference image block are obtained.

[0077] Based on the block matching 3D filtering algorithm, the estimated parameters are used as hyperparameters to construct a denoising model. The RAW format image is input into the denoising model for filtering and the denoised target image is output.

[0078] In one embodiment, when the computer program is executed by a processor, it further performs the following steps: selecting the central region of a parallelogram as a reference image block for Gaussian noise estimation, and calculating the mean of the central region serving as the reference image block:

[0079] in, The mean of the reference image patch, The total number of pixels in the reference image patch. To select the number of sub-bins, for The order, The reference image patch. And the variance:

[0080] in, The variance of the reference image patch is given. The histogram of the reference image patch is obtained. The Gaussian distribution of the central region is estimated based on the mean and variance, and the probability density function is obtained.

[0081] in, Let be the probability density function. The input pixel values ​​are used. The histogram is analyzed and compared with the probability density function. If the comparison results are similar, the mean and variance are used as estimation parameters for the Gaussian noise corresponding to the reference image patch.

[0082] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0083] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0084] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A method for denoising RAW format images, characterized in that, The method includes: Acquire a RAW format image and extract the straight line features from the RAW format image; Based on the straight line features, a parallelogram is determined, and the central region of the parallelogram is selected as a reference image block for Gaussian noise estimation, thereby obtaining the histogram and estimated parameters of the reference image block. Based on the block matching 3D filtering algorithm, the estimated parameters are used as hyperparameters to construct a denoising model. The RAW format image is input into the denoising model for filtering and the denoised target image is output.

2. The method according to claim 1, characterized in that, Acquire a RAW format image and extract the straight line features from the RAW format image, including: A standard color chart corresponding to a RAW format image is acquired using an image sensor. Randomly selected pixels from the RAW format image data are mapped to a polar coordinate plane and then accumulated to obtain an accumulation plane. Based on the accumulated plane, obtain the line corresponding to the local maximum value that meets the threshold requirement, and search for the line features in the Cartesian coordinate system corresponding to the two endpoints of the line.

3. The method according to claim 1, characterized in that, Determining a parallelogram based on the aforementioned straight line characteristics includes: A parallelogram is formed by selecting straight lines whose segment length is equal to the minimum threshold length from the straight line features, and the central region of the parallelogram is determined according to the size of the preset standard pixel block: in, The x-axis coordinates of the center point of the central region. The vertical coordinate of the center point of the central region is y = y. Let be the coordinates of the top left vertex of the parallelogram. Let be the coordinates of the top right vertex of the parallelogram. Let be the coordinates of the lower left vertex of the parallelogram. Let be the coordinates of the lower right vertex of the parallelogram. , , , These are the coordinates of the four vertices of the central region: top left, top right, bottom left, and bottom right. The pixel block size.

4. The method according to claim 3, characterized in that, The central region of the parallelogram is selected as a reference image block for Gaussian noise estimation, resulting in a histogram and estimated parameters for the reference image block. The process of obtaining the histogram and estimated parameters for the reference image block includes: The central region of the parallelogram is selected as a reference image block for Gaussian noise estimation, and the mean of the central region of the reference image block is calculated: in, The mean of the reference image patch, The total number of pixels in the reference image patch. To select the number of sub-blocks, for The order, For reference image blocks; And variance: in, The variance of the reference image patch; Obtain the histogram of the reference image block; Based on the mean and variance, estimate the Gaussian distribution of the central region to obtain the probability density function: in, Let be the probability density function. The input pixel value; The histogram and the probability density function are analyzed and compared. If the comparison results are similar, the mean and the variance are used as estimation parameters of the Gaussian noise corresponding to the reference image patch.

5. The method according to claim 3, characterized in that, Before the step of determining the parallelogram based on the straight line characteristics, the method further includes: The endpoints that make up the line segments in the line feature are deleted from the RAW format image, and the coordinates of each endpoint are recorded.

6. The method according to claim 4, characterized in that, Analyzing and comparing the histogram with the probability density function, if the comparison results are similar, the mean and variance are used as estimation parameters for the Gaussian noise corresponding to the reference image patch, including: The histogram and the probability density function are analyzed and compared. After the comparison results are arranged in order, the mean and variance of the comparison results that are sorted within the standard interval are taken as the estimation parameters of the Gaussian noise corresponding to the reference image block.

7. The method according to any one of claims 4 to 6, characterized in that, Based on the block-matching 3D filtering algorithm, a denoising model is constructed using the estimated parameters as hyperparameters. The RAW format image is input into the denoising model for filtering processing, and the denoised target image is output, including: Based on the block matching 3D filtering algorithm, the estimated parameters are used as hyperparameters to construct a denoising model. The RAW format image is input into the denoising model to complete image block segmentation and similar block search, 3D transformation and filtering, and inverse transformation and aggregation filtering. The image is then enhanced by histogram equalization, and the denoising objective function is output.

8. A noise reduction device for RAW format images, characterized in that, The device includes: A geometric feature acquisition module is used to acquire RAW format images and extract the straight line features of the RAW format images; The Gaussian noise estimation module is used to determine the parallelogram based on the straight line features, select the central region of the parallelogram as a reference image block for Gaussian noise estimation, and obtain the histogram and estimation parameters of the reference image block. The denoising module is used to construct a denoising model based on the block matching 3D filtering algorithm, using the estimated parameters as hyperparameters, inputting the RAW format image into the denoising model for filtering, and outputting a denoised target image.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images