Fluorescence image adaptive enhancement noise reduction method and fluorescence endoscope imaging system

Abstract

The invention discloses a fluorescence image adaptive enhancement noise reduction method and a fluorescence endoscope imaging system, and relates to the image processing related field, and the method comprises the steps: loading a fluorescence endoscope image; dividing the fluorescent endoscope image to obtain D window fluorescent images; extracting a fluorescence image of the d-th window; filtering and noise reduction are carried out to generate a d-th window noise reduction image; performing noise reduction verification to generate a d-th window noise reduction verification image; continuously performing filtering noise reduction and noise reduction verification on each window fluorescence image in the D window fluorescence images to generate D window noise reduction verification images; and carrying out image fusion on the D window noise reduction verification images to generate a fluorescence endoscope noise reduction image. The technical problem that the quality of the fluorescence endoscope noise reduction image is uneven due to the fact that an existing fluorescence endoscope image noise reduction method is not ideal in noise reduction effect and lacks a unified verification process is solved, and the technical effect of improving the quality of the fluorescence endoscope noise reduction image is achieved.

Description

Technical Field

[0001] The present application relates to the field of image processing, and in particular to a method for adaptively enhancing and reducing noise in fluorescence images and a fluorescence endoscope imaging system. Background Art

[0002] Fluorescence navigation endoscope system is a medical device that uses fluorescence technology to assist doctors in diagnosis and surgery. However, due to the insufficient sensitivity of the system, fluorescence imaging often has a low signal-to-noise ratio problem, that is, it is difficult to effectively distinguish the foreground signal from the background noise, which affects the image quality and may reduce the accuracy of the doctor's diagnosis and the precision of the surgical operation. Existing methods mainly focus on image filtering and noise reduction, but often due to the algorithm is not sophisticated or targeted enough, it is difficult to effectively distinguish the foreground signal from the background noise, resulting in unsatisfactory noise reduction effect. At the same time, due to the lack of a unified noise reduction and verification process, the quality of the final noise-reduced image is uneven.

[0003] In the current related technologies, the fluorescence endoscope image denoising method has technical problems such as uneven quality of fluorescence endoscope denoised images due to unsatisfactory denoising effects and lack of a unified verification process. Summary of the invention

[0004] The present application provides a fluorescence image adaptive enhancement denoising method and a fluorescence endoscope imaging system, which achieves the technical effect of improving the quality of fluorescence endoscope denoised images by combining an adaptive enhancement denoising algorithm and an enhancement denoising verification channel to perform refined filtering denoising and verification on each window fluorescence image, and using a Laplace pyramid fusion algorithm to fuse the processed images to generate high-quality fluorescence endoscope denoised images.

[0005] The present application provides a fluorescence image adaptive enhancement and noise reduction method, comprising:

[0006] Loading fluorescence endoscopy images;

[0007] Dividing the fluorescent endoscope image according to preset image windows to obtain D window fluorescent images, where D is the total number of window fluorescent images corresponding to the fluorescent endoscope image;

[0008] Extracting the d-th window fluorescence image according to the D window fluorescence images, wherein d is a positive integer and d belongs to D;

[0009] Performing filtering and noise reduction on the fluorescence image of the dth window according to an adaptive enhanced noise reduction algorithm to generate a noise reduction image of the dth window;

[0010] Performing noise reduction verification on the d-th window noise reduction image according to the enhanced noise reduction verification channel to generate a d-th window noise reduction verification image;

[0011] Based on the adaptive enhanced noise reduction algorithm and the enhanced noise reduction verification channel, continue to filter and reduce noise and verify the noise reduction of each window fluorescence image in the D window fluorescence images, and generate D window noise reduction verification images in combination with the d-th window noise reduction verification image;

[0012] The D window denoising verification images are fused according to the Laplace pyramid fusion algorithm to generate a fluorescence endoscope denoising image.

[0013] In a possible implementation, the fluorescence image of the dth window is filtered and denoised according to an adaptive enhanced denoising algorithm to generate a denoised image of the dth window, and the following processing is performed:

[0014] Perform RGB channel separation according to the dth window fluorescence image to generate a ternary channel image;

[0015] According to the brightness filter function, brightness of each channel image in the ternary channel image is purified respectively to generate a three-channel image brightness layer;

[0016] Based on the brightness layer of the three-channel image, and according to the detail filter function, respectively purifying the details of each channel image to generate a detail layer of the three-channel image;

[0017] Based on the three-channel image brightness layer and the three-channel image detail layer, performing noise reduction optimization according to the adaptive enhanced noise reduction channel to generate a three-channel noise reduction image;

[0018] The RGB channel fusion is performed according to the three-channel denoised image to obtain the d-th window denoised image.

[0019] In a possible implementation, the following processing is performed:

[0020] The brightness filter function is:

[0021] L(x,y)=F guidfilter (I(x,y));

[0022] Among them, L(x,y) represents the brightness layer of the channel image, F guidfilter represents the guided filter function, and I(x,y) represents the channel image.

[0023] In a possible implementation, the following processing is performed:

[0024] The detail filter function is:

[0025] D(x,y)=I(x,y)-L(x,y);

[0026] Among them, D(x,y) represents the channel image detail layer, I(x,y) represents the channel image, and L(x,y) represents the channel image brightness layer.

[0027] In a possible implementation, based on the three-channel image brightness layer and the three-channel image detail layer, noise reduction optimization is performed according to the adaptive enhanced noise reduction channel to generate a three-channel noise reduction image, and the following processing is performed:

[0028] Perform signal-to-noise ratio calculation based on the ternary channel image to obtain a ternary signal-to-noise ratio coefficient;

[0029] The adaptive enhanced noise reduction channel includes a signal-to-noise ratio deviation analysis branch, a noise reduction gain analysis branch and a noise reduction adaptive branch;

[0030] Inputting the ternary signal-to-noise ratio coefficient into the signal-to-noise ratio deviation analysis branch to obtain a ternary signal-to-noise ratio deviation, wherein the signal-to-noise ratio deviation analysis branch includes an endoscopic image signal-to-noise ratio threshold;

[0031] Inputting the ternary signal-to-noise ratio deviation into the noise reduction gain parsing branch to obtain a ternary gain parsing result, wherein the noise reduction gain parsing branch includes a brightness gain parsing table and a detail gain parsing table, and each gain parsing result in the ternary gain parsing result includes a brightness gain coefficient and a detail gain coefficient;

[0032] Based on the brightness layer of the three-channel image, the detail layer of the three-channel image and the ternary gain analysis result, denoising is performed according to the denoising adaptive branch to obtain the three-channel denoised image, wherein the denoising adaptive branch includes a denoising adaptive function, and the denoising adaptive function is:

[0033] NRI(x,y)=LGC*L(x,y)+DGC*D(x,y);

[0034] Among them, NRI(x,y) represents the channel denoised image, LGC represents the brightness gain coefficient, L(x,y) represents the channel image brightness layer, DGC represents the detail gain coefficient, and D(x,y) represents the channel image detail layer.

[0035] In a possible implementation, the denoising verification is performed on the denoised image of the d-th window according to the enhanced denoising verification channel to generate the denoising verification image of the d-th window, and the following processing is performed:

[0036] The enhanced noise reduction verification channel includes an image Weber quality calculation branch and a Weber verification judgment branch;

[0037] Performing a Weber quality analysis on the d-th window denoised image based on the image Weber quality calculation branch to generate a d-th image Weber quality characteristic coefficient;

[0038] Inputting the Weber quality feature coefficient of the dth image into the Weber verification judgment branch, wherein the Weber verification judgment branch includes a preset image Weber quality feature constraint;

[0039] If the Weber quality feature coefficient of the d-th image satisfies the preset image Weber quality feature constraint, obtaining a verification pass instruction, and adding the d-th window denoised image to the d-th window denoised verification image based on the verification pass instruction;

[0040] If the Weber quality feature coefficient of the dth image does not satisfy the preset image Weber quality feature constraint, a verification warning instruction is obtained, and based on the verification warning instruction, the dth window denoised image is deeply optimized to obtain the dth window denoised verification image.

[0041] In a possible implementation, based on the image Weber quality calculation branch, Weber quality analysis is performed on the d-th window denoised image to generate the d-th image Weber quality feature coefficient, and the following processing is performed:

[0042] Perform semantic segmentation according to the d-th window denoised image to generate a d-th background partition image and a d-th target partition image;

[0043] Performing grayscale concentration value calculation based on the dth background partition image to generate a dth background grayscale concentration coefficient;

[0044] Calculate the grayscale concentration value based on the d-th target subarea image to generate the d-th target grayscale concentration coefficient;

[0045] The dth background grayscale concentration coefficient and the dth target grayscale concentration coefficient are input into the image Weber quality calculation branch to obtain the dth image Weber quality characteristic coefficient, wherein the image Weber quality calculation branch includes an image Weber quality calculation function, and the image Weber quality calculation function is:

[0046]

[0047] Among them, IWMSC represents the image Weber quality characteristic coefficient, BGC represents the background grayscale concentration coefficient, and TGC represents the target grayscale concentration coefficient.

[0048] The present application also provides a fluorescent endoscope imaging system, comprising:

[0049] A fluorescent endoscope image loading module, wherein the fluorescent endoscope image loading module is used to load the fluorescent endoscope image;

[0050] D window fluorescence image acquisition modules, the D window fluorescence image acquisition modules are used to divide the fluorescence endoscope image according to preset image windows to obtain D window fluorescence images, where D is the total number of window fluorescence images corresponding to the fluorescence endoscope image;

[0051] A d-th window fluorescence image extraction module, the d-th window fluorescence image extraction module is used to extract the d-th window fluorescence image according to the D window fluorescence images, wherein d is a positive integer and d belongs to D;

[0052] A d-th window noise reduction image generation module, the d-th window noise reduction image generation module is used to filter and reduce noise on the d-th window fluorescence image according to an adaptive enhanced noise reduction algorithm to generate a d-th window noise reduction image;

[0053] A d-th window denoising verification image generation module, the d-th window denoising verification image generation module is used to perform denoising verification on the d-th window denoising image according to the enhanced denoising verification channel to generate a d-th window denoising verification image;

[0054] D window noise reduction verification image generation modules, the D window noise reduction verification image generation modules are used to continue filtering and denoising each window fluorescence image in the D window fluorescence images based on the adaptive enhanced noise reduction algorithm and the enhanced noise reduction verification channel, and generate D window noise reduction verification images in combination with the d-th window noise reduction verification image;

[0055] A fluorescence endoscope denoised image generation module is used to perform image fusion on the D window denoised verification images according to a Laplace pyramid fusion algorithm to generate a fluorescence endoscope denoised image.

[0056] The method for adaptive enhancement and denoising of fluorescence images and the fluorescence endoscope imaging system proposed in the present application are to load a fluorescence endoscope image, divide the fluorescence endoscope image according to a preset image window, obtain D window fluorescence images, extract the d-th window fluorescence image according to the D window fluorescence images, filter and denoise the d-th window fluorescence image according to the adaptive enhancement and denoising algorithm, generate the d-th window denoised image, and then perform denoising verification on the d-th window denoised image according to the enhanced denoising verification channel to generate the d-th window denoised verification image, and then continue to perform filtering and denoising and denoising verification on each window fluorescence image in the D window fluorescence images based on the adaptive enhancement and denoising algorithm and the enhanced denoising verification channel, generate D window denoised verification images by combining the d-th window denoised verification image, and finally perform image fusion on the D window denoised verification images according to the Laplace pyramid fusion algorithm to generate a fluorescence endoscope denoised image, thereby achieving the technical effect of improving the quality of fluorescence endoscope denoised images. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the technical solution of the embodiment of the present invention, the accompanying drawings of the embodiment of the present invention will be briefly introduced below. A flow chart is used in the present application to illustrate the operations performed by the system according to the embodiment of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, various steps can be processed in reverse order or simultaneously as needed. At the same time, other operations can also be added to these processes, or one or more operations can be removed from these processes.

[0058] Figure 1 A schematic diagram of the process of the fluorescence image adaptive enhancement and denoising method provided in an embodiment of the present application;

[0059] Figure 2 A schematic diagram of the structure of a fluorescence endoscope imaging system provided in an embodiment of the present application.

[0060] Explanation of the accompanying drawings: fluorescence endoscope image loading module 10, D window fluorescence image acquisition module 20, d th window fluorescence image extraction module 30, d th window denoised image generation module 40, d th window denoised verification image generation module 50, D window denoised verification image generation module 60, fluorescence endoscope denoised image generation module 70. DETAILED DESCRIPTION

[0061] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below.

[0062] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of this application.

[0063] In the following description, reference is made to "some embodiments", which describe a subset of all possible embodiments, but it is understood that "some embodiments" may be the same subset or different subsets of all possible embodiments, and may be combined with each other without conflict. The terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or modules that are not clearly listed or inherent to these processes, methods, products, or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by technicians in the technical field of this application. The terms used herein are for the purpose of describing the embodiments of the present application only.

[0064] The present application provides a method for adaptively enhancing and reducing noise in fluorescence images, such as Figure 1 As shown, the method includes:

[0065] Step S100, loading the fluorescence endoscope image. Specifically, the fluorescence endoscope imaging system is started and connected to the image processing device, the fluorescence endoscope image file to be loaded is selected or specified on the image processing device, and the fluorescence endoscope image is loaded into the memory of the image processing device to prepare for image processing.

[0066] Step S200, dividing the fluorescence endoscope image according to the preset image window to obtain D window fluorescence images, wherein D is the total number of window fluorescence images corresponding to the fluorescence endoscope image. Specifically, according to the image size and processing requirements, the size and number of the preset image windows are set, the fluorescence endoscope image is divided according to the preset image window size to obtain multiple window fluorescence images, and the total number D of window fluorescence images is calculated and recorded, and the entire image is divided into multiple small windows to improve processing efficiency and accuracy.

[0067] In step S300, the d-th window fluorescence image is extracted according to the D window fluorescence images, wherein d is a positive integer and belongs to D. Specifically, a loop variable d is set with an initial value of 1, and in the loop, the d-th window fluorescence image is extracted from the D window fluorescence images according to the value of d, so as to process each window fluorescence image one by one.

[0068] Step S400, filtering and denoising the fluorescence image of the dth window according to the adaptive enhanced denoising algorithm to generate a denoised image of the dth window. The adaptive enhanced denoising algorithm is a method capable of adaptively adjusting the denoising strategy according to the image content. The adaptive enhanced denoising algorithm automatically estimates the statistical characteristics of the noise by analyzing the input image, and uses these statistical characteristics to suppress or reduce the noise components, while retaining the useful information in the image as much as possible. Specifically, according to the requirements of the adaptive enhanced denoising algorithm, the initialization-related parameters and variables, such as filter coefficients, number of iterations, etc., are determined by experience or experiments, and the fluorescence image of the dth window is analyzed by frequency analysis and other methods to estimate the statistical characteristics of the noise, determine the distribution and intensity of the noise, and apply the adaptive filter to the fluorescence image of the dth window according to the noise characteristics for filtering. The adaptive filter adjusts the pixel value according to the local characteristics of the image by spatial domain filtering or frequency domain filtering and other methods to reduce the influence of noise. After filtering, the denoised denoised image of the dth window is obtained. Through the denoising process, the noise components are suppressed while retaining the main features and detail information of the original image.

[0069] In a possible implementation, step S400 further includes step S410, performing RGB channel separation according to the fluorescence image of the dth window to generate a ternary channel image. Specifically, the RGB pixel data of the fluorescence image of the dth window is read, and the RGB pixel data is separated into three independent channel images of R (red), G (green), and B (blue). Step S420, according to the brightness filter function, the brightness of each channel image in the ternary channel image is purified to generate a three-channel image brightness layer. Specifically, a brightness filter function is designed or selected, and the brightness filter function is applied to the three channel images of R, G, and B respectively, and the brightness information in each channel is extracted to generate a R, G, and B three channel image brightness layer containing the purified brightness information. Step S430, based on the three channel image brightness layer, according to the detail filter function, the details of each channel image are purified to generate a three channel image detail layer. Specifically, a detail filter function is designed or selected, and the three-channel image brightness layer is used as a guide to apply the detail filter function to the three-channel images of R, G, and B, extract the edge, texture and other detail information in the image, and generate the R, G, and B three-channel image detail layer containing the purified detail information. Step S440, based on the three-channel image brightness layer and the three-channel image detail layer, noise reduction optimization is performed according to the adaptive enhanced noise reduction channel to generate a three-channel noise reduction image. Specifically, an adaptive enhanced noise reduction channel is designed or selected, combined with the three-channel image brightness layer and the three-channel image detail layer, and the three-channel R, G, and B channels are subjected to noise reduction processing using the adaptive enhanced noise reduction channel, and the three-channel R, G, and B noise reduction images are generated according to the result of the noise reduction processing. Step S450, RGB channel fusion is performed according to the three-channel noise reduction image to obtain the d-th window noise reduction image. Specifically, the noise reduction images of the three channels of R, G, and B are merged, and the image data of the three channels are recombined into a complete color image through the RGB channel fusion algorithm, and the d-th window noise reduction image is output. This implementation method first separates the RGB channels of the fluorescence image in the dth window and then performs subsequent processing. Since the information in each channel may be affected by different types of noise or have different sensitivities to noise, by separating the channels, targeted noise reduction processing can be performed on each channel. This can more finely control the noise reduction effect and avoid color distortion or detail loss due to over-processing or under-processing of a certain channel, thereby achieving the technical effect of retaining the color characteristics and details of the original image as much as possible.

[0070] In a possible implementation, step S420 further includes step S421, and the brightness filter function is:

[0071] L(x,y)=F guidfilter (I(x,y));

[0072] Among them, L(x,y) represents the brightness layer of the channel image, F guidfilter Represents the guided filter function, I(x,y) represents the channel image. This implementation method uses the guided filter function to extract the brightness layer of the channel image. The guided filter function defines the pixel weights in the filtering process according to the guided image, thereby realizing the purification of the brightness information. In the guided filter, the filter output is the result of the local linear transformation of the input image and the guided image. This local linear relationship ensures that the edge and detail information of the image can be effectively preserved during the filtering process. Therefore, in this way, the technical effect of effectively removing noise and other interference factors while maintaining the edge and details of the image is achieved.

[0073] In a possible implementation, step S430 further includes step S431, and the detail filter function is:

[0074] D(x,y)=I(x,y)-L(x,y);

[0075] Among them, D(x,y) represents the channel image detail layer, I(x,y) represents the channel image, and L(x,y) represents the channel image brightness layer. This implementation method extracts the channel image detail layer by subtracting the channel image brightness layer from the channel image. Since the image brightness layer mainly contains the low-frequency information of the image, that is, the overall brightness of the image and the slowly changing areas, and the image detail layer contains the high-frequency information of the image, that is, the edges, textures and other rapidly changing areas of the image, the low-frequency components in the image can be removed by subtracting the channel image from the channel image brightness layer, thereby achieving the technical effect of highlighting the high-frequency detail information.

[0076] In a possible implementation, step S440 further includes step S441, performing signal-to-noise ratio calculation according to the ternary channel image to obtain a ternary signal-to-noise ratio coefficient. Specifically, the image data of the three channels R, G, and B are read respectively, the signal-to-noise ratio calculation formula is applied to the image data of each channel, the signal-to-noise ratio of each channel is calculated, and the signal-to-noise ratio of the three channels is obtained as the ternary signal-to-noise ratio coefficient. The ternary signal-to-noise ratio coefficient reflects the ratio of signal to noise in the image and describes the intensity and distribution of noise in each channel. Step S442, the adaptive enhanced noise reduction channel includes a signal-to-noise ratio deviation parsing branch, a noise reduction gain parsing branch, and a noise reduction adaptive branch; step S443, the ternary signal-to-noise ratio coefficient is input into the signal-to-noise ratio deviation parsing branch to obtain a ternary signal-to-noise ratio deviation, wherein the signal-to-noise ratio deviation parsing branch includes an endoscopic image signal-to-noise ratio threshold. Specifically, the three branches of the adaptive enhanced noise reduction channel are initialized, and the ternary signal-to-noise ratio coefficient calculated in step S441 is input into the signal-to-noise ratio deviation analysis branch. In the signal-to-noise ratio deviation analysis branch, it is compared with the preset endoscopic image signal-to-noise ratio threshold. According to the comparison result, the signal-to-noise ratio deviation of each channel is calculated. The signal-to-noise ratio deviation refers to the difference between the actual signal-to-noise ratio and the expected or ideal signal-to-noise ratio. By comparing the ternary signal-to-noise ratio coefficient with the preset endoscopic image signal-to-noise ratio threshold, it is used to evaluate the degree of deviation of the signal-to-noise ratio of each channel. Step S444: The ternary signal-to-noise ratio deviation is input into the noise reduction gain analysis branch to obtain a ternary gain analysis result, wherein the noise reduction gain analysis branch includes a brightness gain analysis table and a detail gain analysis table, and each gain analysis result in the ternary gain analysis result includes a brightness gain coefficient and a detail gain coefficient. Specifically, the ternary signal-to-noise ratio deviation calculated in step S443 is input into the noise reduction gain parsing branch, and the noise reduction gain parsing branch determines the gain coefficient that should be applied to each channel in the noise reduction process according to the signal-to-noise ratio deviation, and searches for the corresponding gain coefficient in the brightness gain parsing table and the detail gain parsing table according to the signal-to-noise ratio deviation, and obtains the brightness gain coefficient and detail gain coefficient of each channel as the ternary gain parsing result, and these gain coefficients are used to adjust the weights of the image brightness layer and the image detail layer in the noise reduction process to achieve the best noise reduction effect. Step S445, based on the three-channel image brightness layer, the three-channel image detail layer and the ternary gain parsing result, the noise reduction process is performed according to the noise reduction adaptive branch to obtain the three-channel noise reduction image, wherein the noise reduction adaptive branch includes a noise reduction adaptive function, and the noise reduction adaptive function is:

[0077] NRI(x,y)=LGC*L(x,y)+DGC*D(x,y);

[0078] Among them, NRI (x, y) represents the channel denoised image, LGC represents the brightness gain coefficient, L (x, y) represents the channel image brightness layer, DGC represents the detail gain coefficient, and D (x, y) represents the channel image detail layer. Specifically, the ternary gain analysis result (including the brightness gain coefficient and the detail gain coefficient) obtained in step S444 is matched with the three-channel image brightness layer and the three-channel image detail layer obtained in steps S420 and S430. In the denoising adaptive branch, according to the denoising adaptive function, in combination with the brightness gain coefficient, the channel image brightness layer, the detail gain coefficient and the channel image detail layer, denoising is performed, and the denoising result of each channel is output, that is, the three-channel denoised image. In this implementation, the denoising adaptive branch applies the denoising adaptive function to perform denoising according to the ternary gain analysis result and the brightness layer and detail layer of the three-channel image. The denoising adaptive function comprehensively considers the brightness information and the detail information, and adjusts the denoising intensity according to the signal-to-noise ratio deviation, so as to realize the adaptive adjustment of the denoising strategy according to the characteristics of the image and the noise situation, and achieves the technical effect of obtaining a better denoising effect.

[0079] Step S500, performing noise reduction verification on the d-th window noise reduction image according to the enhanced noise reduction verification channel, and generating the d-th window noise reduction verification image. Specifically, in order to ensure that the noise reduction process is not excessive or insufficient and to ensure the quality of the processed image, the d-th window noise reduction image is quality evaluated using the enhanced noise reduction verification channel, and the enhanced noise reduction verification channel performs a series of quality evaluation operations on the d-th window noise reduction image, such as calculating the image's SNR (signal-to-noise ratio), PSNR (peak signal-to-noise ratio), contrast, clarity and other indicators to quantitatively evaluate the noise reduction effect, and based on the result of the noise reduction quality evaluation, the enhanced noise reduction verification channel generates the d-th window noise reduction verification image.

[0080] In a possible implementation, step S500 further includes step S510, wherein the enhanced noise reduction verification channel includes an image Weber quality calculation branch and a Weber verification judgment branch. Specifically, the image Weber quality calculation branch is used to calculate the Weber quality feature coefficient of the d-th window noise reduction image, and the Weber verification judgment branch is used to determine whether the image meets the quality requirements according to the preset image Weber quality feature constraints. Step S520, based on the image Weber quality calculation branch, Weber quality analysis is performed on the d-th window noise reduction image to generate the d-th image Weber quality feature coefficient. Specifically, the Weber quality analysis is an image quality assessment method based on the human visual system, which obtains an image quality assessment result close to human eye perception by analyzing the image's contrast, brightness, color and other features, inputs the d-th window noise reduction image into the image Weber quality calculation branch, applies the Weber quality analysis algorithm, calculates the image's Weber quality feature coefficient, and outputs the d-th image Weber quality feature coefficient, which reflects the quality level of the image in visual perception. Step S530, input the Weber quality feature coefficient of the dth image into the Weber verification judgment branch, wherein the Weber verification judgment branch includes a preset image Weber quality feature constraint. Specifically, the image Weber quality feature constraint is a set of thresholds or conditions for judging whether the image quality is qualified, and the calculated Weber quality feature coefficient of the dth image is input into the Weber verification judgment branch, and the Weber verification judgment branch judges the input Weber quality feature coefficient according to the preset image Weber quality feature constraint to judge whether the image meets the quality requirement. Step S540, if the Weber quality feature coefficient of the dth image meets the preset image Weber quality feature constraint, obtain a verification pass instruction, and add the dth window denoised image to the dth window denoised verification image based on the verification pass instruction. Specifically, if the Weber quality feature coefficient of the dth image meets the preset constraint condition, it means that the image after denoising has reached the expected quality level in visual perception, so a verification pass instruction is generated, the dth window denoised image is marked as verification passed, and added to the dth window denoised verification image set. Step S550, if the Weber quality feature coefficient of the d-th image does not meet the preset image Weber quality feature constraint, obtain a verification warning instruction, and perform depth optimization on the d-th window denoised image based on the verification warning instruction to obtain the d-th window denoised verification image. Specifically, if the Weber quality feature coefficient of the d-th image does not meet the preset constraint condition, it means that the image after denoising still has problems in visual perception, and it is necessary to further improve the quality through deep optimization. At this time, a verification warning instruction is generated, and according to the verification warning instruction, the d-th window denoised image is subjected to deep optimization processing to improve its quality. The deep optimization may include adjusting the parameters of the denoising algorithm, applying other denoising techniques or combining other image enhancement methods, and using the image after deep optimization as the d-th window denoising verification image.This implementation method uses Weber quality analysis to verify the noise reduction of the d-th window noise reduction image. Since Weber quality analysis uses the visual perception characteristics of the human eye to evaluate image quality, it can more accurately evaluate the quality of the image after noise reduction processing, thereby achieving the technical effect of obtaining more accurate quality evaluation results.

[0081] In a possible implementation, step S520 further includes step S521, performing semantic segmentation according to the dth window denoised image, generating the dth background partition image and the dth target partition image. Specifically, the dth window denoised image is loaded, the semantic segmentation algorithm is applied, the dth window denoised image is divided into two regions, the background and the target, the image of the background region is extracted, the dth background partition image is generated, the image of the target region is extracted, and the dth target partition image is generated. Step S522, grayscale concentration value calculation is performed based on the dth background partition image to generate the dth background grayscale concentration coefficient. Specifically, the dth background partition image is read, the grayscale histogram of the background region is calculated, and the grayscale concentration value, that is, the point with the highest frequency of grayscale value occurrence, is calculated according to the grayscale histogram. The grayscale concentration value reflects the frequency of occurrence of a certain grayscale level in the image, and is an indicator for evaluating the uniformity of the grayscale distribution of the image. The grayscale concentration value is normalized to obtain the dth background grayscale concentration coefficient. Step S523, grayscale concentration value calculation is performed based on the dth target partition image to generate the dth target grayscale concentration coefficient. Specifically, the dth target partition image is read, the grayscale histogram of the target area is calculated, the grayscale concentration value is calculated according to the grayscale histogram, the grayscale concentration value is normalized, and the dth target grayscale concentration coefficient is obtained. The dth target grayscale concentration coefficient reflects the brightness distribution characteristics of the target object and is used to evaluate the visibility and clarity of the target object in the image. Step S524, the dth background grayscale concentration coefficient and the dth target grayscale concentration coefficient are input into the image Weber quality calculation branch to obtain the dth image Weber quality feature coefficient, wherein the image Weber quality calculation branch includes an image Weber quality calculation function, and the image Weber quality calculation function is:

[0082]

[0083] Among them, IWMSC represents the image Weber quality characteristic coefficient, BGC represents the background grayscale concentration coefficient, and TGC represents the target grayscale concentration coefficient. Specifically, the calculated d-th background grayscale concentration coefficient BGC and the d-th target grayscale concentration coefficient TGC are used as input, and the image Weber quality calculation function is applied. The image Weber quality calculation function calculates the image Weber quality characteristic coefficient IWMSC according to BGC and TGC, and outputs the d-th image Weber quality characteristic coefficient IWMSC. This implementation method is based on the analysis of the semantic segmentation and grayscale distribution characteristics of the image. By quantitatively evaluating the brightness changes of the background and target areas, combined with the visual perception characteristics of the human eye, it more accurately evaluates the quality of the image after noise reduction processing, and achieves a technical effect of comprehensively reflecting the quality status of the image.

[0084] Step S600, based on the adaptive enhanced noise reduction algorithm and the enhanced noise reduction verification channel, continue to filter and reduce noise and verify the noise reduction of each window fluorescence image in the D window fluorescence images, and combine the d-th window noise reduction verification image to generate D window noise reduction verification images. Specifically, the loop variable d is incremented until d is equal to D. In each loop, steps S300 to S500 are repeated to filter and reduce noise and verify the noise reduction of each window fluorescence image, thereby obtaining D window noise reduction verification images.

[0085] Step S700, the D window denoising verification images are fused according to the Laplacian pyramid fusion algorithm to generate a fluorescence endoscope denoised image. Specifically, for each window denoising verification image, a corresponding Laplacian pyramid is constructed. The Laplacian pyramid is a multi-scale, multi-resolution image representation method. The original image is decomposed into sub-images of different frequency bands through a series of bandpass filters. Each sub-image reflects the detail information of the original image at different scales. Since different window denoising verification images may have differences in size or spatial position, the Laplacian pyramids of different window denoising verification images are hierarchically aligned before fusion to ensure that different images correspond accurately at each level of the pyramid. A fusion strategy (such as based on pixel value, eigenvalue or other metric) is adopted, starting from the top layer (low-resolution layer) of the Laplace pyramid, and fusion is performed layer by layer. At each layer, the fused pixel value or eigenvalue is calculated according to the fusion strategy to generate the fusion result of the layer. When the fusion of all layers is completed, the fused pyramid layers are merged layer by layer using the reconstruction process of the Laplace pyramid, and finally a fluorescence endoscope denoised image is generated. The Laplace pyramid fusion algorithm can effectively process the scale difference and spatial position difference between different window denoised verification images. By constructing a pyramid and fusion layer by layer, the Laplace pyramid fusion algorithm fully retains the detail information in the original image, while reducing noise interference, thereby generating a high-quality fluorescence endoscope denoised image. The embodiment of the present application adopts a combination of an adaptive enhanced denoising algorithm and an enhanced denoising verification channel to perform refined filtering denoising and verification on each window fluorescence image, and uses a Laplace pyramid fusion algorithm to fuse the processed images to generate a high-quality fluorescence endoscope denoised image and other technical means, thereby achieving the technical effect of improving the quality of the fluorescence endoscope denoised image.

[0086] In the above, refer to Figure 1 The method for adaptive enhancement and noise reduction of fluorescence images according to an embodiment of the present invention is described in detail. Figure 2 A fluorescence endoscopic imaging system according to an embodiment of the present invention is described.

[0087] The fluorescent endoscope imaging system according to the embodiment of the present invention is used to solve the technical problem that the fluorescent endoscope denoising method has uneven quality due to the unsatisfactory denoising effect and lack of a unified verification process, so as to achieve the technical effect of improving the quality of the fluorescent endoscope denoised image. The fluorescent endoscope imaging system includes: a fluorescent endoscope image loading module 10, a D window fluorescent image acquisition module 20, a d-th window fluorescent image extraction module 30, a d-th window denoised image generation module 40, a d-th window denoised verification image generation module 50, a D-th window denoised verification image generation module 60, and a fluorescent endoscope denoised image generation module 70.

[0088] The fluorescence endoscope image loading module 10 is used to load the fluorescence endoscope image;

[0089] The D window fluorescence image acquisition module 20 is used to divide the fluorescence endoscope image according to the preset image windows to obtain D window fluorescence images, where D is the total number of window fluorescence images corresponding to the fluorescence endoscope image;

[0090] The d-th window fluorescence image extraction module 30 is used to extract the d-th window fluorescence image according to the D window fluorescence images, wherein d is a positive integer and d belongs to D;

[0091] The d-th window noise reduction image generation module 40 is used to filter and reduce noise on the d-th window fluorescence image according to an adaptive enhanced noise reduction algorithm to generate a d-th window noise reduction image;

[0092] The d-th window denoising verification image generating module 50 is used to perform denoising verification on the d-th window denoising image according to the enhanced denoising verification channel to generate a d-th window denoising verification image;

[0093] The D window noise reduction verification image generation module 60 is used to continue filtering and noise reduction and noise reduction verification on each window fluorescence image in the D window fluorescence images based on the adaptive enhanced noise reduction algorithm and the enhanced noise reduction verification channel, and generate D window noise reduction verification images in combination with the dth window noise reduction verification image;

[0094] The fluorescence endoscope denoised image generation module 70 is used to perform image fusion on the D window denoised verification images according to the Laplace pyramid fusion algorithm to generate a fluorescence endoscope denoised image.

[0095] The specific configuration of the d-window denoised image generation module 40 will be described in detail below. As described above, the d-window fluorescence image is filtered and denoised according to the adaptive enhanced denoising algorithm to generate the d-window denoised image. The d-window denoised image generation module 40 may further include: an RGB channel separation unit for performing RGB channel separation according to the d-window fluorescence image to generate a ternary channel image; a brightness purification unit for performing brightness purification on each channel image in the ternary channel image according to a brightness filter function to generate a three-channel image brightness layer; a detail purification unit for performing detail purification on each channel image according to a detail filter function based on the three-channel image brightness layer to generate a three-channel image detail layer; a noise reduction optimization unit for performing noise reduction optimization according to the adaptive enhanced denoising channel based on the three-channel image brightness layer and the three-channel image detail layer to generate a three-channel denoised image; and an RGB channel fusion unit for performing RGB channel fusion based on the three-channel denoised image to obtain the d-window denoised image.

[0096] The brightness purification unit may further include: a brightness filter function construction subunit for constructing the brightness filter function as follows:

[0097] L(x,y)=F guidfilter (I(x,y));

[0098] Among them, L(x,y) represents the brightness layer of the channel image, F guidfilter represents the guided filter function, and I(x,y) represents the channel image.

[0099] The detail purification unit may further include: a detail filter function construction subunit for constructing the detail filter function as follows:

[0100] D(x,y)=I(x,y)-L(x,y);

[0101] Among them, D(x,y) represents the channel image detail layer, I(x,y) represents the channel image, and L(x,y) represents the channel image brightness layer.

[0102] Wherein, based on the brightness layer of the three-channel image and the detail layer of the three-channel image, noise reduction optimization is performed according to the adaptive enhanced noise reduction channel to generate a three-channel noise reduction image, and the noise reduction optimization unit may further include: a signal-to-noise ratio calculation subunit is used to perform signal-to-noise ratio calculation according to the ternary channel image to obtain a ternary signal-to-noise ratio coefficient; an adaptive enhanced noise reduction channel construction subunit is used for the adaptive enhanced noise reduction channel to include a signal-to-noise ratio deviation parsing branch, a noise reduction gain parsing branch and a noise reduction adaptive branch; a signal-to-noise ratio deviation parsing subunit is used to input the ternary signal-to-noise ratio coefficient into the signal-to-noise ratio deviation parsing branch to obtain a ternary signal-to-noise ratio deviation, wherein the signal-to-noise ratio deviation parsing branch includes an endoscope. The denoising gain parsing subunit is used to input the ternary signal-to-noise ratio deviation into the denoising gain parsing branch to obtain a ternary gain parsing result, wherein the denoising gain parsing branch includes a brightness gain parsing table and a detail gain parsing table, and each gain parsing result in the ternary gain parsing result includes a brightness gain coefficient and a detail gain coefficient; the denoising processing subunit is used to perform denoising processing according to the denoising adaptive branch based on the three-channel image brightness layer, the three-channel image detail layer and the ternary gain parsing result to obtain the three-channel denoised image, wherein the denoising adaptive branch includes a denoising adaptive function, and the denoising adaptive function is:

[0103] NRI(x,y)=LGC*L(x,y)+DGC*D(x,y);

[0104] Among them, NRI(x,y) represents the channel denoised image, LGC represents the brightness gain coefficient, L(x,y) represents the channel image brightness layer, DGC represents the detail gain coefficient, and D(x,y) represents the channel image detail layer.

[0105] The specific configuration of the d-th window denoising verification image generation module 50 will be described in detail below. As described above, the d-th window denoising image is subjected to denoising verification according to the enhanced denoising verification channel to generate the d-th window denoising verification image. The d-th window denoising verification image generation module 50 may further include: an enhanced denoising verification channel construction unit for the enhanced denoising verification channel including an image Weber quality calculation branch and a Weber verification judgment branch; a Weber quality analysis unit for performing Weber quality analysis on the d-th window denoising image based on the image Weber quality calculation branch to generate the d-th image Weber quality feature coefficient; a Weber verification judgment unit for inputting the d-th image Weber quality feature coefficient into the Weber verification judgment branch, which In the method, the Weber verification judgment branch includes a preset image Weber quality feature constraint; a verification pass instruction acquisition unit is used to obtain a verification pass instruction if the Weber quality feature coefficient of the d-th image satisfies the preset image Weber quality feature constraint, and add the d-th window denoised image to the d-th window denoised verification image based on the verification pass instruction; a verification warning instruction acquisition unit is used to obtain a verification warning instruction if the Weber quality feature coefficient of the d-th image does not satisfy the preset image Weber quality feature constraint, and perform deep optimization on the d-th window denoised image based on the verification warning instruction to obtain the d-th window denoised verification image.

[0106] Wherein, based on the image Weber quality calculation branch, the Weber quality analysis is performed on the dth window denoised image to generate the dth image Weber quality feature coefficient. The Weber quality analysis unit may further include: a semantic segmentation subunit for performing semantic segmentation based on the dth window denoised image to generate the dth background partition image and the dth target partition image; a grayscale concentration value calculation subunit for performing grayscale concentration value calculation based on the dth background partition image to generate the dth background grayscale concentration coefficient, and performing grayscale concentration value calculation based on the dth target partition image to generate the dth target grayscale concentration coefficient; a dth image Weber quality feature coefficient acquisition subunit for inputting the dth background grayscale concentration coefficient and the dth target grayscale concentration coefficient into the image Weber quality calculation branch to obtain the dth image Weber quality feature coefficient, wherein the image Weber quality calculation branch includes an image Weber quality calculation function, and the image Weber quality calculation function is:

[0107]

[0108] Among them, IWMSC represents the image Weber quality characteristic coefficient, BGC represents the background grayscale concentration coefficient, and TGC represents the target grayscale concentration coefficient.

[0109] The fluorescence endoscope imaging system provided in the embodiment of the present invention can execute the fluorescence image adaptive enhancement and denoising method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0110] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules may be used and run on the user terminal and / or server, and the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention.

[0111] The above specific implementation manner does not constitute a limitation to the protection scope of the present application. It should be understood by those skilled in the art that various modifications, combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application should be included in the protection scope of the present application. In some cases, the actions or steps recorded in the present application can be performed in an order different from that in the embodiment and can still achieve the desired results. In addition, the process depicted in the accompanying drawings does not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. A fluorescence image adaptive enhancement and denoising method, characterized in that: The method comprises: Loading fluorescence endoscopy images; Dividing the fluorescent endoscope image according to preset image windows to obtain D window fluorescent images, where D is the total number of window fluorescent images corresponding to the fluorescent endoscope image; Extracting the d-th window fluorescence image according to the D window fluorescence images, wherein d is a positive integer and d belongs to D; Performing filtering and noise reduction on the fluorescence image of the dth window according to an adaptive enhanced noise reduction algorithm to generate a noise reduction image of the dth window; Performing noise reduction verification on the d-th window noise reduction image according to the enhanced noise reduction verification channel to generate a d-th window noise reduction verification image; Based on the adaptive enhanced noise reduction algorithm and the enhanced noise reduction verification channel, continue to filter and reduce noise and verify the noise reduction of each window fluorescence image in the D window fluorescence images, and generate D window noise reduction verification images in combination with the d-th window noise reduction verification image; The D window denoising verification images are fused according to the Laplace pyramid fusion algorithm to generate a fluorescence endoscope denoising image.

2. The method for adaptive enhancement and denoising of fluorescence images according to claim 1, characterized in that: Filtering and denoising the fluorescence image of the dth window according to an adaptive enhanced denoising algorithm to generate a denoised image of the dth window includes: Perform RGB channel separation according to the dth window fluorescence image to generate a ternary channel image; According to the brightness filter function, brightness of each channel image in the ternary channel image is purified respectively to generate a three-channel image brightness layer; Based on the brightness layer of the three-channel image, and according to the detail filter function, respectively purifying the details of each channel image to generate a detail layer of the three-channel image; Based on the three-channel image brightness layer and the three-channel image detail layer, performing noise reduction optimization according to the adaptive enhanced noise reduction channel to generate a three-channel noise reduction image; The RGB channel fusion is performed according to the three-channel denoised image to obtain the d-th window denoised image.

3. The method for adaptive enhancement and denoising of fluorescence images according to claim 2, characterized in that: The brightness filter function is: L(x,y)=F guidfilter (I(x,y)); Among them, L(x, y) represents the brightness layer of the channel image, F guidfilter represents the guided filter function, and I(x, y) represents the channel image.

4. The method for adaptive enhancement and denoising of fluorescence images according to claim 2, characterized in that: The detail filter function is: D(x,y)=I(x,y)-L(x,y); Among them, D(x, y) represents the channel image detail layer, I(x, y) represents the channel image, and L(x, y) represents the channel image brightness layer.

5. The method for adaptive enhancement and denoising of fluorescence images according to claim 2, characterized in that: Based on the three-channel image brightness layer and the three-channel image detail layer, performing noise reduction optimization according to the adaptive enhanced noise reduction channel to generate a three-channel noise reduction image, including: Perform signal-to-noise ratio calculation based on the ternary channel image to obtain a ternary signal-to-noise ratio coefficient; The adaptive enhanced noise reduction channel includes a signal-to-noise ratio deviation analysis branch, a noise reduction gain analysis branch and a noise reduction adaptive branch; Inputting the ternary signal-to-noise ratio coefficient into the signal-to-noise ratio deviation analysis branch to obtain a ternary signal-to-noise ratio deviation, wherein the signal-to-noise ratio deviation analysis branch includes an endoscopic image signal-to-noise ratio threshold; Inputting the ternary signal-to-noise ratio deviation into the noise reduction gain parsing branch to obtain a ternary gain parsing result, wherein the noise reduction gain parsing branch includes a brightness gain parsing table and a detail gain parsing table, and each gain parsing result in the ternary gain parsing result includes a brightness gain coefficient and a detail gain coefficient; Based on the brightness layer of the three-channel image, the detail layer of the three-channel image and the ternary gain analysis result, denoising is performed according to the denoising adaptive branch to obtain the three-channel denoised image, wherein the denoising adaptive branch includes a denoising adaptive function, and the denoising adaptive function is: NRI(x,y)=LGC*L(x,y)+DGC*D(x,y); Among them, NRI(x, y) represents the channel denoised image, LGC represents the brightness gain coefficient, L(x, y) represents the channel image brightness layer, DGC represents the detail gain coefficient, and D(x, y) represents the channel image detail layer.

6. The method for adaptive enhancement and denoising of fluorescence images according to claim 1, characterized in that: Performing noise reduction verification on the noise reduction image of the d-th window according to the enhanced noise reduction verification channel to generate the noise reduction verification image of the d-th window includes: The enhanced noise reduction verification channel includes an image Weber quality calculation branch and a Weber verification judgment branch; Performing a Weber quality analysis on the d-th window denoised image based on the image Weber quality calculation branch to generate a d-th image Weber quality characteristic coefficient; Inputting the Weber quality feature coefficient of the dth image into the Weber verification judgment branch, wherein the Weber verification judgment branch includes a preset image Weber quality feature constraint; If the Weber quality feature coefficient of the d-th image satisfies the preset image Weber quality feature constraint, obtaining a verification pass instruction, and adding the d-th window denoised image to the d-th window denoised verification image based on the verification pass instruction; If the Weber quality feature coefficient of the dth image does not satisfy the preset image Weber quality feature constraint, a verification warning instruction is obtained, and based on the verification warning instruction, the dth window denoised image is deeply optimized to obtain the dth window denoised verification image.

7. The method for adaptive enhancement and denoising of fluorescence images according to claim 6, characterized in that: Performing Weber quality analysis on the d-th window denoised image based on the image Weber quality calculation branch to generate a d-th image Weber quality feature coefficient includes: Perform semantic segmentation according to the d-th window denoised image to generate a d-th background partition image and a d-th target partition image; Performing grayscale concentration value calculation based on the dth background partition image to generate a dth background grayscale concentration coefficient; Calculate the grayscale concentration value based on the d-th target subarea image to generate the d-th target grayscale concentration coefficient; The dth background grayscale concentration coefficient and the dth target grayscale concentration coefficient are input into the image Weber quality calculation branch to obtain the dth image Weber quality characteristic coefficient, wherein the image Weber quality calculation branch includes an image Weber quality calculation function, and the image Weber quality calculation function is: Among them, IWMSC represents the image Weber quality characteristic coefficient, BGC represents the background grayscale concentration coefficient, and TGC represents the target grayscale concentration coefficient.

8. A fluorescence endoscope imaging system, characterized in that: The system is used to implement the fluorescence image adaptive enhancement and denoising method according to any one of claims 1 to 7, and the system comprises: A fluorescent endoscope image loading module, wherein the fluorescent endoscope image loading module is used to load the fluorescent endoscope image; D window fluorescence image acquisition modules, the D window fluorescence image acquisition modules are used to divide the fluorescence endoscope image according to preset image windows to obtain D window fluorescence images, where D is the total number of window fluorescence images corresponding to the fluorescence endoscope image; A d-th window fluorescence image extraction module, the d-th window fluorescence image extraction module is used to extract the d-th window fluorescence image according to the D window fluorescence images, wherein d is a positive integer and d belongs to D; A d-th window noise reduction image generation module, the d-th window noise reduction image generation module is used to filter and reduce noise on the d-th window fluorescence image according to an adaptive enhanced noise reduction algorithm to generate a d-th window noise reduction image; A d-th window denoising verification image generation module, the d-th window denoising verification image generation module is used to perform denoising verification on the d-th window denoising image according to the enhanced denoising verification channel to generate a d-th window denoising verification image; D window noise reduction verification image generation modules, the D window noise reduction verification image generation modules are used to continue filtering and denoising each window fluorescence image in the D window fluorescence images based on the adaptive enhanced noise reduction algorithm and the enhanced noise reduction verification channel, and generate D window noise reduction verification images in combination with the d-th window noise reduction verification image; A fluorescence endoscope denoised image generation module is used to perform image fusion on the D window denoised verification images according to a Laplace pyramid fusion algorithm to generate a fluorescence endoscope denoised image.

Images