Intraoral endoscope image enhancement method, system, medium, and apparatus

Abstract

This application provides a method, system, medium, and device for enhancing images of oral endoscopes. The method includes: acquiring the original image of an oral endoscope; performing illumination enhancement processing on the original image of the oral endoscope using a preset multi-scale Retinex algorithm; inputting the enhanced image into a preset U-Net semantic segmentation network for semantic segmentation processing to determine a binarized oral cyst mask image; performing target region extraction processing on the binarized oral cyst mask image to determine the target region image; performing image enhancement processing on the target region image using a preset super-resolution ESRGAN model to determine the enhanced image of the target region; and superimposing the enhanced image of the target region with the enhanced image according to a preset superposition intensity to determine the enhanced oral endoscope image. This application improves the clarity and diagnostic accuracy of oral cyst region recognition under conditions of low illumination, uneven illumination, and complex tissue backgrounds.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and more specifically, to a method, system, medium, and device for enhancing images of an oral endoscope. Background Technology

[0002] With the continuous development of minimally invasive medicine and digital diagnosis and treatment, endoscopic technology is increasingly widely used in the field of oral medicine, including departments such as stomatology, otolaryngology, and gastroenterology. In oral medicine, it plays a crucial role, particularly in the early screening, preoperative assessment, intraoperative navigation, and postoperative follow-up of oral cysts. Oral cysts, as a common pathological condition, are characterized by their varied morphology and complex distribution. Failure to identify or accurately locate them in a timely manner can lead to serious consequences such as infection, bone destruction, and even malignant transformation. The quality of endoscopic images plays a decisive role in the physician's judgment and treatment strategy. High-quality images not only help physicians accurately identify the morphology, boundaries, and location of cysts but also assist in assessing the degree of compression or invasiveness of the lesion on surrounding tissues. Therefore, obtaining high-quality, structurally clear endoscopic images is essential for physician diagnosis.

[0003] However, due to the complexity of oral anatomy, such as narrow palate, soft tissue folds, and light and shadow effects, images acquired by traditional endoscopes often have the following problems: 1. Uneven lighting and shadow occlusion: Due to the limited space in the oral cavity and the single angle of illumination from the endoscope light source, some areas of the image are overexposed or underexposed, especially in areas near the root of the gums and mucosal folds, where severe brightness gradients are easily formed, affecting tissue recognition.

[0004] 2. Low tissue contrast and blurred boundaries: The difference in color and texture between cysts and surrounding healthy soft tissue is very subtle. Conventional images cannot effectively separate them in terms of grayscale distribution, and the edge details are blurred. Doctors need to rely on experience to make judgments, which can easily lead to missed detections or misdiagnoses.

[0005] 3. Insufficient image resolution and detail fidelity: Limited by the hardware capabilities of endoscopic devices, especially portable devices, the low image sampling frequency and limited pixel density result in a lack of clear expression of lesion microstructures (such as cyst walls and exudative areas).

[0006] 4. Traditional image enhancement algorithms have limitations: Methods such as histogram equalization, CLAHE, and LIME are mainly based on global or local brightness redistribution, lacking an adaptive processing mechanism for lesion areas. They are prone to side effects such as over-enhancement, loss of detail, and color distortion, and cannot meet the actual diagnostic needs.

[0007] 5. Lack of integrated hardware and software solutions: Current image enhancement mainly relies on post-processing software, which does not form a closed loop with the acquisition, analysis, and display processes, and cannot meet the needs of dentists for "real-time enhancement and immediate diagnosis" in clinical practice.

[0008] In practical applications, intraoral endoscopic images are affected by factors such as unstable lighting environment, complex tissue structure, strong reflection on mucosal surface, and limited visible area, resulting in low image quality and seriously affecting the doctor's clinical judgment.

[0009] Therefore, there is an urgent need for an image enhancement solution specifically designed for the intraoral endoscopic environment. This solution should not only be able to correct for non-uniform lighting, but also have the ability to intelligently identify and restore the structure of lesions. Furthermore, it should be deployable in embedded medical devices to support real-time processing, helping doctors improve the efficiency and accuracy of cyst identification. Summary of the Invention

[0010] In view of the deficiencies in the prior art, the purpose of this application is to provide a method, system, medium and device for enhancing intraoral endoscopic images.

[0011] A first aspect of this application provides a method for enhancing intraoral endoscopic images, comprising: Obtain raw images from an oral endoscope; The original image of the oral endoscope is subjected to illumination enhancement processing using a preset multi-scale Retinex algorithm to determine the enhanced image; The enhanced image is input into a preset U-Net semantic segmentation network for semantic segmentation processing to determine a binarized oral cyst mask image; The binarized oral cyst mask image is subjected to target region extraction processing to determine the target region image, which represents the oral cyst region image; The target region image is enhanced using a preset super-resolution ESRGAN model to determine the enhanced image of the target region. Based on a preset overlay intensity, the enhanced image of the target region is overlaid with the enhanced image to determine the enhanced intraoral endoscopic image.

[0012] Optionally, the step of performing illumination enhancement processing on the original image of the oral endoscope using a preset multi-scale Retinex algorithm to determine the enhanced image includes: The original image from the oral endoscope is modeled as a product of the reflection and illumination components, and the expression for the original image of the oral endoscope with respect to the reflection and illumination components is determined as follows: Take the logarithm of the expressions for the reflection and illumination components of the original image from the oral endoscope to determine the logarithmic expression for the reflection component: Based on the Gaussian convolution kernel at each scale and the logarithmic expression of the reflection component, the reflection component at each scale is fused at multiple scales to determine the multi-scale fused image. The enhanced image is determined by performing color restoration on the multi-scale fused image using a preset color restoration function.

[0013] Optionally, the preset U-Net semantic segmentation network includes an encoder, a decoder, and skip connections; The step of inputting the enhanced image into a preset U-Net semantic segmentation network for semantic segmentation processing to determine the mask map of the oral cyst region includes: The enhanced image is input into the encoder for multi-scale semantic information extraction to determine downsampling features; The downsampled features are input into the decoder for upsampling to determine the upsampled features; The downsampled features and the upsampled features are concatenated through the skip connection to determine the mask image of the oral cyst region; Based on a preset threshold hyperparameter, the mask image of the oral cyst region is thresholded and binarized to determine the binarized oral cyst mask image.

[0014] Optionally, the preset super-resolution ESRGAN model includes a backbone generator network, an upsampling module, and a discriminator module; The step of performing image enhancement processing on the target region image using a preset super-resolution ESRGAN model to determine the enhanced target region image includes: The target region image is input into the backbone generator network for feature extraction to determine the feature image of the oral cyst region. The feature image of the oral cyst region is input into the upsampling module for upsampling, and the discriminator module is used to determine the image type to identify the enhanced image of the target region.

[0015] Optionally, the step of superimposing the enhanced image of the target region with the enhanced image according to a preset superposition intensity to determine the enhanced intraoral endoscopic image includes: The enhanced image of the target region is input into a preset edge detection module to determine the edge amplitude map of the enhanced image of the target region. Threshold binarization is performed on the edge magnitude map of the enhanced image of the target region to determine the binary edge mask map of the enhanced image of the target region; Based on the preset superposition intensity, the binary edge mask image of the enhanced image of the target region is superimposed with the enhanced image to determine the enhanced intraoral endoscope image.

[0016] A second aspect of this application provides an image enhancement system for an oral endoscope, comprising: The image acquisition module is used to acquire raw images from the oral endoscope; The illumination enhancement module is used to perform illumination enhancement processing on the original image of the oral endoscope using a preset multi-scale Retinex algorithm to determine the enhanced image; The semantic segmentation module is used to input the enhanced image into a preset U-Net semantic segmentation network for semantic segmentation processing to determine a binarized oral cyst mask image; The target region extraction module is used to perform target region extraction processing on the binarized oral cyst mask image to determine the target region image, wherein the target region image represents the oral cyst region image. The image enhancement module is used to perform image enhancement processing on the target region image using a preset super-resolution ESRGAN model, and to determine the enhanced image of the target region; The image display module is used to overlay the enhanced image of the target region with the enhanced image according to a preset overlay intensity to determine the enhanced intraoral endoscopic image.

[0017] A third aspect of this application provides a non-transitory computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the steps of any of the methods provided in the first aspect of this application.

[0018] A fourth aspect of this application provides an electronic device comprising: A memory on which computer programs are stored; A processor for executing the computer program in the memory to implement the steps of any of the methods provided in the first aspect of this application.

[0019] This application discloses an image enhancement method for oral endoscopy, which employs a pre-defined multi-scale Retinex algorithm for illumination enhancement, a pre-defined U-Net semantic segmentation network for lesion region segmentation, and a pre-defined super-resolution ESRGAN model for target region image enhancement. This method effectively improves the clarity and diagnostic accuracy of oral cysts under complex background conditions such as low illumination, uneven illumination, low tissue contrast, and blurred edges, thereby increasing the efficiency of oral cyst lesion identification.

[0020] Other technical effects resulting from the additional features will be further illustrated in the corresponding embodiments. Attached Figure Description

[0021] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a flowchart illustrating an intraoral endoscopic image enhancement method according to an exemplary embodiment.

[0022] Figure 2 This is a schematic diagram of the structure of an oral endoscope image enhancement system according to an exemplary embodiment. Detailed Implementation

[0023] The present application will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present application, but do not limit the present application in any way. It should be noted that those skilled in the art can make several modifications and improvements without departing from the concept of the present application. These all fall within the protection scope of the present application.

[0024] Traditional image enhancement techniques in the prior art, such as histogram equalization and contrast stretching, can only improve the overall brightness or local sharpness of an image, making it difficult to simultaneously maintain detail fidelity and color naturalness in oral cyst areas. Existing medical image segmentation algorithms still rely on high-quality labeled images and lack robustness and adaptability to low illumination. Furthermore, the application of super-resolution algorithms in the medical field still faces problems such as algorithm complexity, time-consuming inference, and difficulty in embedding into terminal devices, and has not yet achieved engineering-scale closed-loop integration. Based on the above problems, this embodiment provides an image enhancement method for oral endoscopy to solve the aforementioned issues.

[0025] Figure 1 This is a flowchart illustrating an intraoral endoscopic image enhancement method according to an exemplary embodiment.

[0026] Reference Figure 1 As shown in one embodiment of this application, a method for enhancing intraoral endoscopic images is provided, including steps S11 to S16.

[0027] S11, acquire the raw image of the oral endoscope.

[0028] S12, the original image of the oral endoscope is enhanced by using the preset multi-scale Retinex algorithm to determine the enhanced image.

[0029] Specifically, the preset multi-scale Retinex algorithm refers to the improved multi-scale Retinex algorithm (Multi-Scale Retinex with Color Restoration, MSRCR).

[0030] S13, the enhanced image input is processed by the preset U-Net semantic segmentation network to determine the binarized oral cyst mask image.

[0031] S14, perform target region extraction processing on the binarized oral cyst mask image to determine the target region image.

[0032] Specifically, the target region represents the oral cyst region, and the target region image represents the oral cyst region image.

[0033] S15, the target region image is enhanced using a preset super-resolution ESRGAN model to determine the enhanced image of the target region.

[0034] S16, according to the preset superposition intensity, the enhanced image of the target area is superimposed on the enhanced image to determine the enhanced intraoral endoscope image.

[0035] The embodiments described above employ a pre-defined multi-scale Retinex algorithm to enhance the illumination of the original images from the oral endoscope, correcting uneven lighting and improving image grayscale balance and tissue contrast. A pre-defined U-Net semantic segmentation network is used for semantic segmentation, accurately segmenting lesion regions and improving the accuracy of oral cyst region localization. A pre-defined super-resolution ESRGAN model is used to enhance the image of the oral cyst region, improving detail representation and effectively improving the emotional appeal and diagnostic accuracy of oral cyst recognition under complex background conditions such as low illumination, uneven lighting, low tissue contrast, and blurred transformations, thereby increasing the efficiency of oral cyst lesion identification.

[0036] To acquire the raw images from the oral endoscope, in some specific embodiments of this application, for S11, acquiring the raw images from the oral endoscope can be done using a professional oral endoscope to capture raw images at 1080p resolution. and the original image It is transmitted to the image processing chip in standard RGB format.

[0037] Specifically, the acquisition frequency of the original images from the oral endoscope can be 10 to 30 frames per second. An LED point light source is set at the front end of the lens of the oral endoscope, and the LED point light source supports multi-angle uniform illumination.

[0038] In order to enhance the illumination of the original image of the oral endoscope, in some specific embodiments of this application, for S12, a preset multi-scale Retinex algorithm is used to perform illumination enhancement processing on the original image of the oral endoscope to determine the enhanced image, which can be done by the following S121 to S124.

[0039] S121, Model the original image of the oral endoscope as a product of the reflection component and the illumination component, and determine the expression of the original image of the oral endoscope with respect to the reflection component and the illumination component.

[0040] Specifically, the pre-defined multi-scale Retinex algorithm (MSRCR) is used to model the image as a product of the reflection and illumination components based on the local adaptive features of the human eye.

[0041] The original image from the oral endoscope is expressed with respect to the reflection and illumination components as follows: in, This represents the original image from an oral endoscope. Represents the reflection component. Indicates the lighting component.

[0042] S122, Take the logarithm of the expression for the reflection component and illumination component of the original image from the oral endoscope to determine the logarithmic expression for the reflection component.

[0043] Specifically, the logarithmic expression for the reflection component is: in, This represents the Gaussian convolution kernel.

[0044] Gaussian convolution kernel Used to represent illumination estimation functions at different scales: in, This represents the control parameters of the illumination estimation function.

[0045] S123, based on the Gaussian convolution kernel and the logarithmic expression of the reflection component at each scale, the reflection component at each scale is fused at multiple scales to determine the multi-scale fused image.

[0046] Specifically, in, Represents a multi-scale fused image. This represents the weight of the i-th reflection component. The scale is represented as Gaussian convolution kernel.

[0047] Specifically, the process of multi-scale fusion of the reflection components at each scale is represented as using multi-scale Retinex, that is, using different filters to smooth and differ the image multiple times, and finally weighted fusion.

[0048] S124, use a preset color restoration function to perform color restoration on the multi-scale fused image and determine the enhanced image.

[0049] Specifically, in, This represents the image pixel values ​​of the enhanced image. This represents the adjustment coefficient. Indicates the scaling factor. This indicates the control parameters.

[0050] Specifically, the color preset's color restoration function is used to enhance the naturalness of the image; the adjustment coefficient... Used to balance the weights of the original image and the enhanced image; scaling factor. Used to control the dynamic range of image pixel values ​​for enhancing images.

[0051] The enhanced image is represented as: in, Indicates an enhanced image. This represents the image pixel values ​​of the enhanced image. This represents the gain adjustment coefficient. This indicates a brightness offset.

[0052] Specifically, enhance the image pixel values ​​of the image. Includes Retinex enhancement and color correction information; It can also represent contrast; brightness offset. Used to adjust the overall brightness of the image.

[0053] The embodiments described above employ a multi-scale Retinex algorithm to perform a multi-scale feature fusion illumination compensation strategy on the original images of the oral endoscope, thereby achieving dark area compensation and bright area suppression, effectively improving image brightness balance and tissue contrast. A preset color restoration function is also used to effectively enhance the naturalness of the image.

[0054] In some possible embodiments, the deep neural network used to perform semantic segmentation processing employs a preset U-Net semantic segmentation network, which includes an encoder, a decoder, and skip connections. The enhanced image is input into the preset U-Net semantic segmentation network to extract the oral cyst region. Other types of deep neural networks may also be used.

[0055] Specifically, the encoder consists of multiple convolutional layers and max-pooling layers to perform downsampling operations; the decoder performs upsampling operations and, in conjunction with skip connections, recovers spatial details.

[0056] To achieve segmentation of the oral cyst lesion area, in some specific embodiments of this application, for S13, the enhanced image is input into a preset U-Net semantic segmentation network for semantic segmentation processing to determine a binarized oral cyst mask image, which can be implemented using S131 to S134.

[0057] S131, the enhanced image input encoder is used to extract multi-scale semantic information and determine downsampling features.

[0058] Specifically: in, ReLU represents the downsampled features, and ReLU() represents the activation function. Indicates to Layer by layer Convolution operation.

[0059] S132, input the downsampled features into the decoder for upsampling operation to determine the upsampled features.

[0060] S133, the downsampled features and upsampled features are concatenated through skip connections to determine the mask map of the oral cyst region.

[0061] Based on steps S132 and S133, the downsampled features and upsampled features are concatenated through skip connections, and then passed through an activation function to obtain a mask image of the oral cyst region: in, A mask representing the oral cyst region, where ReLU represents the activation function. This indicates a splicing operation. This indicates an upsampling operation.

[0062] Specifically, the mask image of the oral cyst region is also represented as the pixel classification result; stitching operation This indicates merging features from different layers along the channel dimension; upsampling operation. This indicates the restoration of spatial resolution and detail information.

[0063] in, This represents the predicted probability of an oral cyst at a pixel in the oral cyst region.

[0064] S134, according to the preset threshold hyperparameter, perform threshold binarization on the mask image of the oral cyst region to determine the binarized oral cyst mask image.

[0065] Specifically, the preset threshold hyperparameter is set as follows: The binarized mask image of oral cysts is represented as follows: ; in, This represents the predicted probability of an oral cyst for each pixel in the oral cyst region. This represents a binarized mask image of an oral cyst. This represents the preset threshold hyperparameter.

[0066] Specifically, the oral cyst prediction probability of each pixel in the oral cyst region output by the preset U-Net semantic segmentation network is compared with the preset threshold hyperparameter, so that pixels with a prediction probability not lower than the threshold hyperparameter are marked as oral cysts (1), and pixels with a probability lower than the threshold hyperparameter are marked as non-oral cysts (0), thereby generating a binarized oral cyst mask map.

[0067] In the above embodiments of this application, a preset U-Net semantic segmentation network is used to extract the oral cyst region, perform multi-scale feature extraction and skip connections, and generate a binarized oral cyst mask map to improve the accuracy of lesion target region extraction.

[0068] In some specific embodiments of this application, the preset super-resolution ESRGAN (Enhanced Super-Resolution Generative Adversarial Network) model includes a backbone generator network, an upsampling module, and a discriminator module. The preset super-resolution ESRGAN model is used for binary mask images of oral cyst regions. The original image sub-regions are magnified and details are restored.

[0069] In order to enhance the oral cyst region, in some specific embodiments of this application, for S15, a preset super-resolution ESRGAN model is used to perform image enhancement processing on the oral cyst region to determine the target region enhancement image, which can be done as follows: S151 to S152.

[0070] S151, the target region image is input into the backbone generator network for feature extraction processing to determine the feature image of the oral cyst region.

[0071] S152, the feature image of the oral cyst region is input into the upsampling module for upsampling operation, and the discriminator module is used to perform image type judgment operation to determine the target region enhancement image.

[0072] Specifically, the backbone generator network comprises multiple residual-in-dense blocks (RDBs). The backbone generator network constructs multiple residual-in-dense blocks based on the input oral cyst region image, and the calculation formula is as follows: in, Represents a residual dense block. Image showing the area of ​​an oral cyst. Indicates a densely connected structure. This represents the residual weight.

[0073] Each residual dense block (RDB) contains multiple convolutional layers, ReLU activation functions, and dense connections, forming a strong expressive power for features. Furthermore, the local dense connections between layers enable information fusion at multiple scales, as shown below: in, This represents the nonlinear transformation of the i-th residual dense block. This is the output of the previous residual dense block.

[0074] Multiple residual dense blocks can be stacked to build up deeper feature learning capabilities.

[0075] After passing through multiple residual dense blocks, the feature image of the oral cyst region is output. This feature image is then input into the upsampling module to restore the low-resolution feature image to a high-resolution version, achieving 2x and 4x magnification, as shown below: in, This represents the oral cyst region image output by the last residual dense block. Upsample() represents the upsampling operation. This indicates an enhanced image of the target region.

[0076] The discriminator module D is used to perform image type judgment operations, distinguishing between the generated "generated high-resolution image" and the "real high-resolution image". In this embodiment, the discriminator can be a PatchGAN discriminator.

[0077] Through adversarial learning, the backbone generator network outputs more realistic details. The loss function of the discriminator module D includes adversarial loss. in, Indicating resistance to loss, Represents a true high-resolution image. This represents the generated high-resolution image; Traditional pixel mean square error often fails to reflect perceptual realism. The pre-defined super-resolution ESRGAN introduces perceptual loss, first extracting deep features from the generated and real images through the VGG network, then calculating L2 loss in the feature space, and finally achieving a generated high-resolution image that is closer to a real high-resolution image in high-level perceptual semantics. in, Indicates perceived loss. This represents the characteristics of a network layer.

[0078] By using perceptual loss, we can avoid simply pursuing pixel overlap and ignoring texture details; Finally, the enhanced image of the target region after being magnified by the upsampling module. Record .

[0079] In the above embodiments of this application, a preset super-resolution ESRGAN model is used to reconstruct the image framework and enhance the image of the oral cyst region, thereby achieving edge sharpening, texture restoration and spatial super-resolution improvement of the oral cyst region.

[0080] In order to determine the enhanced intraoral endoscopic image, in some specific embodiments of this application, for S16, the enhanced image of the target area is superimposed on the enhanced image according to a preset superposition intensity to determine the enhanced intraoral endoscopic image, which can be done as follows: S161 to S163.

[0081] S161, Input the enhanced image of the target region into the preset edge detection module to determine the edge amplitude map of the enhanced image of the target region.

[0082] Specifically, the edge magnitude map of the enhanced image of the target region is represented as follows: .

[0083] S162, perform threshold binarization on the edge magnitude map of the enhanced image of the target region to determine the binary edge mask map of the enhanced image of the target region.

[0084] Specifically, the binary edge mask map of the enhanced image of the target region is represented as follows: .

[0085] S163, according to the preset superposition intensity, the binary edge mask image of the target region enhancement image is superimposed with the enhancement image to determine the enhanced oral endoscope image.

[0086] Specifically, the overlay process is as follows: in, This indicates an enhanced intraoral endoscopic image. Indicates an enhanced image. This indicates the preset superposition strength. A binary edge mask representing the enhanced image of the target region.

[0087] The enhanced intraoral endoscopic images are then fed back into the backbone generator network to optimize the generated high-resolution images using edge information. Structural details and boundary fidelity.

[0088] The embodiments described above in this application effectively enhance the oral cyst region in the oral endoscopy image by superimposing a binary edge mask image of the target region onto the enhanced image, thereby facilitating the precise localization of the oral cyst region.

[0089] This application provides a method for enhancing oral endoscopy images. It employs a multi-scale Retinex algorithm for illumination enhancement, a pre-defined U-Net semantic segmentation network for lesion region segmentation, a pre-defined super-resolution ESRGAN model for image enhancement, and superimposes enhanced images of the target region onto the enhanced images to obtain the final enhanced oral endoscopy image. This method achieves excellent image enhancement results and is applicable to multiple stages such as initial screening of oral cysts, preoperative navigation, and postoperative follow-up. It can also be extended to other endoscopic medical imaging scenarios, providing strong support for improving diagnostic efficiency and accuracy.

[0090] The image enhancement method for oral endoscopy provided in this application can be widely used in multiple oral specialties such as oral surgery, endodontics, and maxillofacial surgery, and can also be extended to other clinical endoscopic scenarios such as otolaryngology endoscopy and gastrointestinal endoscopy.

[0091] The preferred features in the above embodiments can be used individually in any embodiment, or in any combination thereof, provided they do not conflict with each other. Furthermore, parts not described in detail in the embodiments can be implemented using existing technologies.

[0092] Figure 2 This is a schematic diagram of the structure of an oral endoscope image enhancement system according to an exemplary embodiment.

[0093] Reference Figure 2 As shown in one embodiment of this application, an oral endoscope image enhancement system 100 is provided, including: an image acquisition module 110, an illumination enhancement module 120, a semantic segmentation module 130, a target region extraction module 140, an image enhancement module 150, and an image display module 160.

[0094] Image acquisition module 110 is used to acquire raw images from an oral endoscope; The illumination enhancement module 120 is used to perform illumination enhancement processing on the original image of the oral endoscope using a preset multi-scale Retinex algorithm to determine the enhanced image; The semantic segmentation module 130 is used to input the enhanced image into a preset U-Net semantic segmentation network for semantic segmentation processing to determine a binarized oral cyst mask image; The target region extraction module 140 is used to perform target region extraction processing on the binarized oral cyst mask image to determine the target region image, which represents the oral cyst region image. Image enhancement module 150 is used to perform image enhancement processing on the oral cyst region using a preset super-resolution ESRGAN model to determine the enhanced image of the target region; The image display module 160 is used to overlay the enhanced image of the target area with the enhanced image according to a preset overlay intensity to determine the enhanced intraoral endoscope image.

[0095] In the embodiments described above, the illumination enhancement module 120 uses a preset multi-scale Retinex algorithm to enhance the illumination of the original image of the oral endoscope, corrects uneven illumination, and enhances image grayscale balance and tissue contrast. The semantic segmentation module 130 uses a preset U-Net semantic segmentation network to perform semantic segmentation, accurately segmenting the lesion area and improving the accuracy of oral cyst location. The image enhancement module 150 uses a preset super-resolution ESRGAN model to enhance the image of the oral cyst area, improve the detail expression, and effectively improve the clarity and diagnostic accuracy of oral cyst identification under complex background conditions such as low illumination, uneven illumination, low tissue contrast, and blurred transformations, thereby improving the efficiency of oral cyst lesion identification.

[0096] The oral endoscope image enhancement system provided in this application has the advantages of strong real-time performance, high target recognition accuracy, and strong structural restoration ability. It is suitable for various scenarios such as clinical oral cyst examination, intraoperative navigation and postoperative image analysis, and supports deployment in embedded medical terminals to meet real-time processing requirements.

[0097] As shown in Table 1, in some possible embodiments, the hardware integration of an intraoral endoscopic image enhancement system also includes a CMOS lens, a Jetson chip, a light source control, a display terminal, and a communication module: Table 1 As shown in Table 1, the image display module 160 can also display enhanced intraoral endoscopy images in real time on the display terminal, and support doctors to interactively analyze and export records. Doctors can perform operations such as zooming, annotation, and export on the display terminal. Images can also be uploaded to a remote cloud platform via Wi-Fi or 5G to support expert collaborative diagnosis.

[0098] The intraoral endoscopic image enhancement system described in the above embodiments of this application can be integrated into an embedded intelligent terminal platform, possesses edge computing capabilities, supports closed-loop operation of image acquisition, processing, and display, and is suitable for real-time applications in clinical diagnostic and treatment environments. The functional modules are interconnected via a communication bus, and the overall system structure is modularly designed, supporting remote maintenance and software upgrades.

[0099] Regarding the embodiments of the above-mentioned oral endoscope image enhancement system, the specific methods of operation of each module can be referred to the description of the corresponding steps of the oral endoscope image enhancement method in any of the above embodiments, and will not be described in detail here.

[0100] Based on the same technical concept, in some specific embodiments of this application, a terminal includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it can be used to perform the intraoral endoscopic image enhancement method described in any of the above embodiments.

[0101] Based on the same technical concept, in some specific embodiments of this application, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, can be used to perform the intraoral endoscopic image enhancement method described in any of the above embodiments.

[0102] Optionally, an electronic device includes a memory and a processor. The memory stores programs; it may include volatile memory, such as random-access memory (RAM), such as static random-access memory (SRAM), double data rate synchronous dynamic random-access memory (DDR SDRAM), etc.; the memory may also include non-volatile memory, such as flash memory. The memory stores computer programs (such as application programs, functional modules, etc. that implement the above methods), computer instructions, etc., which may be partitioned and stored in one or more memories. Furthermore, the computer programs, computer instructions, data, etc., can be accessed by the processor.

[0103] The aforementioned computer programs, computer instructions, etc., can be stored in partitions within one or more memory locations. Furthermore, the aforementioned computer programs, computer instructions, data, etc., can be accessed by a processor.

[0104] A processor is used to execute a computer program stored in memory to implement the various steps of the methods involved in the above embodiments. For details, please refer to the relevant descriptions in the preceding method embodiments.

[0105] The processor and memory can be separate structures or integrated structures. When the processor and memory are separate structures, they can be coupled together via a bus.

[0106] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0107] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0108] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0109] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0110] Alternatively, a non-transitory computer-readable storage medium may be used, on which a computer program is stored.

[0111] The foregoing has described some specific embodiments of this application. It should be understood that this application is not limited to the specific embodiments described above, and those skilled in the art can make various modifications or variations within the scope of the claims, which do not affect the substantive content of this application. The above-described preferred features can be used in any combination without conflict.

Claims

1. An intraoral endoscope image enhancement method, characterized by, include: Acquire raw images from an oral endoscope; The original image of the oral endoscope is subjected to illumination enhancement processing using a preset multi-scale Retinex algorithm to determine the enhanced image; The enhanced image is input into a preset U-Net semantic segmentation network for semantic segmentation processing to determine a binarized oral cyst mask image; The binarized oral cyst mask image is subjected to target region extraction processing to determine the target region image, which represents the oral cyst region image; The target region image is enhanced using a preset super-resolution ESRGAN model to determine the enhanced image of the target region. Based on a preset overlay intensity, the enhanced image of the target region is overlaid with the enhanced image to determine the enhanced intraoral endoscopic image.

2. The oral endoscope image enhancing method according to claim 1, characterized by, The process of using a preset multi-scale Retinex algorithm to perform illumination enhancement processing on the original image of the oral endoscope and determining the enhanced image includes: The original image from the oral endoscope is modeled as a product of the reflection and illumination components, and the expression for the original image from the oral endoscope with respect to the reflection and illumination components is determined as follows: Take the logarithm of the expressions for the reflection and illumination components of the original image from the oral endoscope to determine the logarithmic expression for the reflection component: Based on the Gaussian convolution kernel at each scale and the logarithmic expression of the reflection component, the reflection component at each scale is fused at multiple scales to determine the multi-scale fused image. The enhanced image is determined by performing color restoration on the multi-scale fused image using a preset color restoration function.

3. The method for enhancing intraoral endoscopic images according to claim 2, characterized in that, The original image from the oral endoscope is expressed as follows with respect to the reflection and illumination components: ； in, This represents the original image from the oral endoscope. This represents the reflection component. Indicates the lighting component; The logarithmic expression for the reflection component is: ； in, Represents the Gaussian convolution kernel; The step of fusing the reflection components at each scale according to the Gaussian convolution kernel at each scale and the logarithmic expression of the reflection components to determine the multi-scale fused image includes: ； ； in, This refers to the multi-scale fused image. This represents the weight of the reflection component described in the i-th term. The scale is represented as Gaussian convolution kernel; The step of performing color restoration on the multi-scale fused image using a preset color restoration function to determine the enhanced image includes: ； in, This represents the image pixel values ​​of the enhanced image. This represents the adjustment coefficient. Indicates the scaling factor. Indicates control parameters; ； in, This refers to the enhanced image. This represents the image pixel values ​​of the enhanced image. This represents the gain adjustment coefficient. This indicates the brightness offset.

4. The method for enhancing intraoral endoscopic images according to claim 1, characterized in that, The preset U-Net semantic segmentation network includes an encoder, a decoder, and skip connections; The step of inputting the enhanced image into a preset U-Net semantic segmentation network for semantic segmentation processing to determine a binarized oral cyst mask image includes: The enhanced image is input into the encoder for multi-scale semantic information extraction to determine downsampling features; The downsampled features are input into the decoder for upsampling to determine the upsampled features; The downsampled features and the upsampled features are concatenated through the skip connection to determine the mask image of the oral cyst region; Based on a preset threshold hyperparameter, the mask image of the oral cyst region is thresholded and binarized to determine the binarized oral cyst mask image.

5. The method for enhancing intraoral endoscopic images according to claim 4, characterized in that, The step of inputting the enhanced image into the encoder for multi-scale semantic information extraction and determining downsampled features includes: ； in, Represents the downsampling feature, and ReLU represents the activation function. ) indicates to Layer by layer Convolution operation; The step of concatenating the downsampled features and the upsampled features through the skip connection to determine the mask image of the oral cyst region includes: ； in, This represents a mask image of the oral cyst region, where ReLU represents the activation function. This indicates a splicing operation. Indicates an upsampling operation; ； Among them, the This represents the predicted probability of an oral cyst for each pixel in the oral cyst region. The step of threshold binarizing the mask image of the oral cyst region according to a preset threshold hyperparameter, and determining the binarized oral cyst mask image, includes: ； ； in, This represents the predicted probability of an oral cyst for each pixel in the oral cyst region. This represents the binarized oral cyst mask image. This represents the preset threshold hyperparameter.

6. The method for enhancing intraoral endoscopic images according to claim 1, characterized in that, The preset super-resolution ESRGAN model includes a backbone generator network, an upsampling module, and a discriminator module; The step of performing image enhancement processing on the target region image using a preset super-resolution ESRGAN model to determine the enhanced target region image includes: The target region image is input into the backbone generator network for feature extraction to determine the feature image of the oral cyst region. The feature image of the oral cyst region is input into the upsampling module for upsampling, and the discriminator module is used to determine the image type to identify the enhanced image of the target region.

7. The method for enhancing intraoral endoscopic images according to claim 1, characterized in that, The step of superimposing the enhanced image of the target region with the enhanced image according to a preset superposition intensity to determine the enhanced intraoral endoscopic image includes: The enhanced image of the target region is input into a preset edge detection module to determine the edge amplitude map of the enhanced image of the target region. Threshold binarization is performed on the edge magnitude map of the enhanced image of the target region to determine the binary edge mask map of the enhanced image of the target region; Based on the preset superposition intensity, the binary edge mask image of the enhanced image of the target region is superimposed with the enhanced image to determine the enhanced intraoral endoscope image.

8. An image enhancement system for oral endoscopes, characterized in that, include: The image acquisition module is used to acquire raw images from the oral endoscope; The illumination enhancement module is used to perform illumination enhancement processing on the original image of the oral endoscope using a preset multi-scale Retinex algorithm to determine the enhanced image; The semantic segmentation module is used to input the enhanced image into a preset U-Net semantic segmentation network for semantic segmentation processing to determine a binarized oral cyst mask image; The target region extraction module is used to perform target region extraction processing on the binarized oral cyst mask image to determine the target region image, wherein the target region image represents the oral cyst region image. The image enhancement module is used to perform image enhancement processing on the target region image using a preset super-resolution ESRGAN model, and to determine the enhanced image of the target region; The image display module is used to overlay the enhanced image of the target region with the enhanced image according to a preset overlay intensity to determine the enhanced intraoral endoscopic image.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps of the method described in any one of claims 1-7.

10. An electronic device, characterized in that, include: A memory on which computer programs are stored; A processor for executing the computer program in the memory to implement the steps of the method according to any one of claims 1-7.

Images