A fiber endoscope image lesion detection method and system

Abstract

The present invention relates to the technical field of image lesion detection, and in particular to a fiber endoscope image lesion detection method and system, the method comprising the following steps: providing a data quality guidance interface; capturing narrowband light images at a first preset position and a second preset position of the endoscope away from the intestinal wall during axial movement, calculating the brightness gain values ​​of the blue light channel and the green light channel, calculating the asymmetric scattering correction coefficient, and updating the indication of the completeness of the color information calibration; tracking the pixel movement speed of interference objects in the field of view during the advancement operation, calculating the relative depth of the interference objects, and updating the indication of the validity of the interference layer depth data; identifying image areas not blocked by interference objects, and updating the indication of the background area information coverage; triggering an image synthesis operation; performing color compensation on the image information of the image areas not blocked by interference objects stored in the background canvas; reconstructing a clear image; and performing lesion detection on the clear image. The method improves the accuracy of lesion detection.

Description

Technical Field

[0001] The present invention relates to the technical field of image lesion detection, and in particular to a fiber endoscope image lesion detection method and system. Background Art

[0002] Acquiring clear and accurate mucosal images is crucial for fiberoptic endoscopy in the examination of early, minimal lesions in the digestive tract. To enhance the field of view and remove deposits, fluid flushing is often performed through the endoscope. However, this procedure can easily cause the fluid to mix with body fluids and debris, creating a turbid environment.

[0003] For example, during a detailed examination of a specific intestinal area, if the patient's bowel preparation is incomplete or local irrigation is performed to remove debris, a layer of turbid liquid of varying thickness forms between the optical lens at the front of the endoscope and the target intestinal wall. Light from the endoscope's light source must pass through this liquid layer on its way to the intestinal wall. This light is absorbed and scattered by suspended particles within it, significantly attenuating the energy of the light ultimately reflected back to the imaging sensor. This results in a generally grayish, hazy appearance in the image, significantly reducing overall contrast and clarity. Summary of the Invention

[0004] The purpose of the present invention is to address the above-mentioned deficiencies and provide a fiber endoscope image lesion detection method and system.

[0005] The present invention adopts the following technical solutions:

[0006] A method for detecting lesions in fiber endoscope images, the method comprising the following steps: S1: providing a data quality guidance interface for indicating the completeness of color information calibration, the validity of interference layer depth data, and the coverage of background area information; S2: responding to the axial movement of the endoscope, capturing narrowband light images of the endoscope at a first preset position and a second preset position away from the intestinal wall during the axial movement, calculating the brightness gain values ​​of the blue light channel and the green light channel based on the narrowband light images at the first preset position and the second preset position, calculating the asymmetric scattering correction coefficient based on the brightness gain values ​​of the blue light channel and the green light channel, and updating the indication of the completeness of the color information calibration based on the stability of the asymmetric scattering correction coefficient; S3: responding to the advancement operation of the endoscope, tracking the pixel movement speed of the interference object in the field of view during the advancement operation, and based on the pixel movement speed of the interference object, Calculate the relative depth of the interference object to obtain the interference layer depth distribution map, and update the indication of the validity of the interference layer depth data according to the validity of the interference layer depth distribution map; S4: respond to the scanning operation of the endoscope, identify the image area that is not blocked by the interference object, store the image information of the image area that is not blocked by the interference object to the background canvas, and update the indication of the background area information coverage according to the coverage of the background canvas; S5: trigger the image synthesis operation when the indicated color information calibration completeness, interference layer depth data validity and background area information coverage have all been achieved; S6: perform color compensation on the image information of the image area that is not blocked by the interference object stored in the background canvas according to the asymmetric scattering correction coefficient; S7: combine the interference layer depth distribution map and the background canvas to reconstruct a clear image; S8: perform lesion detection on the clear image.

[0007] Through the above solution, by actively collecting image information during the endoscopic operation and combining color correction, depth estimation and background construction, the problems of image blur, dynamic artifacts and color distortion in turbid liquid environments are effectively overcome, thereby reconstructing clear and color-accurate images, significantly improving the accuracy of lesion detection.

[0008] To further solve the problem, the present application further proposes that step S2 includes: responding to the axial movement of the endoscope, continuously capturing an image sequence for color correction during the axial movement; dividing each frame of the image sequence for color correction into a central area and an edge area; and respectively calculating the brightness changes of the blue light channel and the green light channel in the central area and the brightness changes of the blue light channel and the green light channel in the edge area;

[0009] Compare the brightness changes of the blue light channel in the central area with those in the edge areas, and compare the brightness changes of the green light channel in the central area with those in the edge areas to identify whether there is a particle crowding effect in the central area;

[0010] The brightness change of the blue light channel in the edge area is used as the true brightness change of the blue light channel reflecting the inherent optical attenuation characteristics under the body fluid interference environment, and the brightness change of the green light channel in the edge area is used as the true brightness change of the green light channel reflecting the inherent optical attenuation characteristics under the body fluid interference environment;

[0011] Calculate the brightness gain value of the blue light channel according to the actual brightness change of the blue light channel;

[0012] Calculate the brightness gain value of the green light channel according to the actual brightness change of the green light channel;

[0013] Calculate the asymmetric scattering correction coefficient according to the brightness gain values ​​of the blue light channel and the green light channel;

[0014] The indication of the completeness of the color information calibration is updated based on the stability of the asymmetric scatter correction coefficient.

[0015] Through the above scheme, by dividing the core area and the edge area and comparing the brightness changes, the particle exclusion effect can be identified, and the gain value and correction coefficient can be calculated based on the actual brightness change of the edge area, so as to perform color correction more accurately and improve the completeness of color information calibration.

[0016] To improve the solution, this application also proposes that some steps in step S3 include:

[0017] In response to a propulsion operation of the endoscope, continuously capturing an image sequence for depth estimation during the propulsion operation;

[0018] From the image sequence used for depth estimation, identify the image region where the distractor is located in the field of view;

[0019] For the image area where the interference object is located, the motion vector of the pixel point in the image area where the interference object is located is estimated between consecutive frames of the image sequence used for depth estimation;

[0020] According to the motion vector of the pixel point, the pixel movement speed of the interference object is calculated.

[0021] Through the above scheme, by continuously capturing image sequences and estimating the motion vectors of the interferer pixels, the pixel movement speed of the interferer can be calculated more accurately, providing reliable data for subsequent depth estimation.

[0022] To improve the solution, this application also proposes that some steps in step S3 include:

[0023] In response to the advancement operation of the endoscope, a compound detection action is performed, wherein the compound detection action includes a uniform advancement phase and a momentary pause phase;

[0024] In the instantaneous pause stage, the particles in the interference are classified into first inertia particles and second inertia particles according to the motion velocity attenuation characteristics of the particles in the interference;

[0025] In the uniform advancing stage, the relative depth of the first inertial particle is calculated according to the pixel moving speed of the first inertial particle;

[0026] According to the relative depth of the first inertial particle, the distance to the intestinal mucosa is determined;

[0027] According to the distance to the intestinal mucosa, the spatial area between the endoscope tip and the intestinal mucosa is determined as the depth range of the fluid layer;

[0028] Setting the relative depth of the second inertial particle to a preset depth or depth range within the depth range of the fluid layer;

[0029] Integrating the relative depth information of the first inertial particles and the relative depth information of the second inertial particles to generate an interference layer depth distribution map;

[0030] The indication of validity of the interference layer depth data is updated according to the validity of the interference layer depth distribution map.

[0031] Through the above scheme, by performing a composite detection action and classifying the interference objects according to the attenuation characteristics of the particle motion velocity, it is possible to distinguish interference objects of different properties and calculate their relative depths respectively, thereby more accurately generating the interference layer depth distribution map and improving the validity of the depth data.

[0032] To improve the solution, this application also proposes that step S4 includes:

[0033] In response to a scanning operation of the endoscope, continuously capturing an image sequence for background area accumulation during the scanning operation;

[0034] For each frame of the image sequence used for background area accumulation, a local area that meets the definition requirement in the frame is identified to obtain a clear area;

[0035] Evaluate the optical uniformity of each clear area to quantify its residual interference level and generate corresponding clarity quality parameters;

[0036] The image information of each clear area and its clarity quality parameters are stored together;

[0037] Perform image registration on the image information of the clear area, and accumulate the registered image information to the background canvas;

[0038] During the accumulation of the registered image information, the image information corresponding to the clear area position stored in the background canvas is compared with the image information stored at the corresponding position, and based on the values ​​of their respective clarity quality parameters, the image information with the higher clarity quality parameter value is selected for updating or merging into the background canvas;

[0039] The coverage of the background canvas is calculated according to the ratio of the number of valid pixels filled in the background canvas to the total number of pixels, and the indication of the coverage of the background area information is updated.

[0040] Through the above scheme, by identifying local areas that meet the clarity requirements and evaluating their optical uniformity, combined with image registration and image information selection and fusion based on clarity quality parameters, the background canvas can be accumulated and updated more effectively to ensure the coverage and quality of background area information.

[0041] To improve the solution, this application also proposes that the steps of evaluating the optical uniformity of each clear area to quantify its residual interference level include:

[0042] Divide the clear area into multiple sub-areas;

[0043] Capture the image sequence corresponding to the clear area;

[0044] For each sub-region in the image sequence corresponding to the clear area, the brightness change, color change and texture change of the pixels in each sub-region are calculated to evaluate the optical uniformity of each sub-region;

[0045] Generate corresponding residual interference metrics based on the brightness changes, color changes, and texture changes of pixels in each sub-region;

[0046] Based on the residual interference metrics of all sub-regions, the residual interference degree of the clear area is comprehensively quantified.

[0047] Through the above scheme, by dividing the clear area into sub-areas and calculating the brightness, color and texture changes of the pixels in each sub-area, the degree of residual interference can be evaluated and quantified more finely, providing more accurate quality parameters for the construction of the background canvas.

[0048] To improve the solution, the present application further proposes that, for each sub-region in the image sequence corresponding to the clear region, the steps of calculating the brightness change, color change, and texture change of the pixels in each sub-region include:

[0049] Obtain the brightness information of the pixels in the sub-region in the time dimension, and calculate the brightness fluctuation amplitude of the pixels in the sub-region based on the brightness information in the time dimension;

[0050] Obtaining the color information of the pixels in the sub-region in the time dimension, and calculating the color fluctuation amplitude of the pixels in the sub-region based on the color information in the time dimension;

[0051] The texture information of the pixels in the sub-region in the time dimension is obtained, and the texture fluctuation amplitude of the pixels in the sub-region is calculated based on the texture information in the time dimension.

[0052] Through the above scheme, by obtaining the brightness, color and texture information of pixels in the sub-region in the time dimension and calculating its fluctuation amplitude, the optical uniformity of the sub-region can be comprehensively and quantitatively reflected, providing a detailed basis for the generation of residual interference metrics.

[0053] To improve the solution, the present application further proposes that the steps of generating a corresponding residual interference metric based on the brightness change, color change, and texture change of pixels in each sub-region include:

[0054] According to the brightness fluctuation amplitude, color fluctuation amplitude and texture fluctuation amplitude of the pixels in the sub-region, weighted fusion is performed to obtain the residual interference measure of the sub-region.

[0055] Through the above scheme, by performing weighted fusion on the brightness, color and texture fluctuation amplitude of pixels in the sub-region, the residual interference degree of the sub-region can be comprehensively evaluated to obtain a more representative residual interference metric.

[0056] To improve the solution, the present application further proposes that the steps of performing weighted fusion based on the brightness fluctuation amplitude, color fluctuation amplitude, and texture fluctuation amplitude of the pixels in the sub-region to obtain the residual interference measurement of the sub-region include:

[0057] Determine the weights corresponding to the brightness fluctuation amplitude, color fluctuation amplitude, and texture fluctuation amplitude of the pixels in the sub-region based on preset rules or dynamic adjustment rules;

[0058] According to the brightness fluctuation amplitude, color fluctuation amplitude, texture fluctuation amplitude and corresponding weights of the pixels in the sub-region, weighted fusion is performed to obtain the residual interference measure of the sub-region.

[0059] Through the above scheme, by determining weights based on preset or dynamically adjusted rules, the brightness, color and texture fluctuation amplitudes are weightedly fused, making the calculation of the residual interference metric more flexible and accurate, and adapting to different interference situations.

[0060] To improve the solution, the present application also proposes a fiber endoscope image lesion detection system, which is applied to the above-mentioned fiber endoscope image lesion detection method. The system includes:

[0061] The data quality guidance module is used to provide a data quality guidance interface and indicate the completeness of color information calibration, the validity of interference layer depth data, and the coverage of background area information;

[0062] a color correction module, configured to respond to axial movement of the endoscope, capture narrowband light images at a first preset position and a second preset position of the endoscope away from the intestinal wall during the axial movement, calculate brightness gain values ​​of a blue light channel and a green light channel based on the narrowband light images at the first preset position and the second preset position, calculate an asymmetric scattering correction coefficient based on the brightness gain values ​​of the blue light channel and the green light channel, and update an indication of color information calibration completeness based on a degree of stability of the asymmetric scattering correction coefficient;

[0063] a depth estimation module, configured to respond to the advancement operation of the endoscope, track the pixel movement speed of the interference object in the field of view during the advancement operation, calculate the relative depth of the interference object based on the pixel movement speed of the interference object, obtain an interference layer depth distribution map, and update the indication of the validity of the interference layer depth data based on the validity of the interference layer depth distribution map;

[0064] a background construction module, configured to respond to a scanning operation of the endoscope, identify an image area not blocked by an interfering object, store image information of the image area not blocked by the interfering object into a background canvas, and update an indication of background area information coverage according to coverage of the background canvas;

[0065] An image synthesis module is configured to trigger an image synthesis operation when the indications of color information calibration completeness, interference layer depth data validity, and background area information coverage are all met, and perform color compensation on the image information of the image area not blocked by the interference object stored in the background canvas according to the asymmetric scattering correction coefficient, and reconstruct a clear image by combining the interference layer depth distribution map and the background canvas;

[0066] The lesion detection module is used to perform lesion detection on clear images.

[0067] Through the above scheme, a system is provided to implement the above-mentioned fiber endoscope image lesion detection method. Through modular design, the entire detection process is made more integrated and automated, thereby improving operational efficiency and detection reliability.

[0068] To further understand the features and technical contents of the present invention, please refer to the following detailed description and drawings of the present invention. However, the drawings provided are only for reference and illustration and are not intended to limit the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 This is a flow chart of a method for detecting lesions in fiber endoscope images according to the present invention;

[0070] Figure 2 The figure is a schematic structural diagram of a fiber endoscope image lesion detection system of the present invention. DETAILED DESCRIPTION

[0071] The following is an explanation of the embodiments of the present invention through specific embodiments. Those skilled in the art can understand the advantages and effects of the present invention from the contents disclosed in this specification. The present invention can be implemented or applied through other different specific embodiments, and the details in this specification can also be modified and changed based on different viewpoints and applications without departing from the spirit of the present invention. In addition, the drawings of the present invention are only for simple schematic illustrations and are not depicted according to actual dimensions. It is stated in advance. The following embodiments will further explain the relevant technical contents of the present invention in detail, but the disclosed contents are not intended to limit the scope of protection of the present invention.

[0072] This embodiment provides a fiber endoscope image lesion detection method and system, combined with Figure 1 and Figure 2 shown.

[0073] refer to Figure 1 , this application proposes a fiber endoscope image lesion detection method, the method comprising the following steps:

[0074] S1: Provides a data quality guidance interface, which is used to indicate the completeness of color information calibration, the validity of interference layer depth data, and the coverage of background area information; S2: Responds to the axial movement of the endoscope, captures the narrow-band light images of the first preset position and the second preset position of the endoscope away from the intestinal wall during the axial movement, calculates the brightness gain values ​​of the blue light channel and the green light channel based on the narrow-band light images of the first preset position and the second preset position, calculates the asymmetric scattering correction coefficient based on the brightness gain values ​​of the blue light channel and the green light channel, and updates the indication of the completeness of color information calibration based on the stability of the asymmetric scattering correction coefficient; S3: Responds to the advancement operation of the endoscope, tracks the pixel movement speed of the interference object in the field of view during the advancement operation, calculates the relative depth of the interference object based on the pixel movement speed of the interference object, and obtains Interference layer depth distribution map, and according to the validity of the interference layer depth distribution map, update the indication of the validity of the interference layer depth data; S4: respond to the scanning operation of the endoscope, identify the image area that is not blocked by the interference object, store the image information of the image area that is not blocked by the interference object to the background canvas, and update the indication of the background area information coverage according to the coverage of the background canvas; S5: when the indicated color information calibration completeness, interference layer depth data validity and background area information coverage have been achieved, trigger the image synthesis operation; S6: according to the asymmetric scattering correction coefficient, perform color compensation on the image information of the image area that is not blocked by the interference object stored in the background canvas; S7: combine the interference layer depth distribution map and the background canvas to reconstruct a clear image; S8: perform lesion detection on the clear image.

[0075] The data quality guidance interface refers to a visual human-computer interaction interface, which can be implemented using a graphical user interface (GUI) or a real-time data display panel, such as a progress bar, color coding, or numerical indicator on a monitor. Its primary purpose is to provide the operator with real-time feedback on the current image data quality, guiding the operator to adjust the endoscope's operating strategy to obtain data that meets subsequent processing requirements. Color information calibration completeness is a metric used to assess whether the data required for image color correction is sufficient and reliable. It can be quantified using statistical methods or machine learning models, such as by the convergence or fluctuation range of the asymmetric scatter correction coefficient. It is primarily used to ensure the accuracy of subsequent color compensation and avoid color distortion caused by insufficient or unstable data. Interference layer depth data validity is a metric used to assess whether the acquired interference object depth information is accurate and comprehensive. It can be determined based on the completeness, continuity, or match of the depth map with the actual scene, such as by analyzing the pixel fill rate or the rationality of the depth value distribution of the interference layer depth distribution map. It is primarily used to provide reliable spatial depth information for subsequent image reconstruction and effectively remove the influence of interference objects. Background area information coverage is a metric used to assess the accumulation of clear image information within the background canvas. It can be calculated based on the ratio of the number of filled valid pixels to the total number of pixels, for example, by calculating the percentage of non-empty pixels in the background canvas. Its primary purpose is to ensure that the reconstructed background image possesses sufficient integrity and detail, providing a comprehensive baseline image for lesion detection. Axial movement refers to the movement of the endoscope along its longitudinal axis. This can be demonstrated by changes in the distance between the endoscope tip and the intestinal wall, such as when the endoscope moves forward toward or backward away from the intestinal wall. This is primarily used to capture images at different optical path lengths to analyze the attenuation and scattering characteristics of light in turbid media. Narrowband light images are images captured using a light source with a specific wavelength range. These can be acquired using an LED light source or filters with specific wavelengths, such as imaging using only blue or green light. Their primary purpose is to enhance the contrast of blood vessels in the mucosal surface, aiding in lesion identification. The asymmetric scattering correction coefficient refers to a parameter used to quantify and compensate for the differences in the degree of scattering of light of different wavelengths in turbid media. It can be calculated using a ratio or functional relationship based on the brightness gain values ​​of the two channels, for example, by deriving it through the brightness change trend of the blue and green light channels. Its main purpose is to correct the image color distortion caused by scattering and restore the true color of the image. The propulsion operation refers to the movement of the endoscope forward along its long axis into the intestine. It can be reflected by the continuous advancement of the front end of the endoscope in the intestine, such as straight or curved propulsion in the intestine. Its main purpose is to track the dynamic behavior of interference objects in the field of view during the movement of the endoscope to estimate their relative depth.The pixel movement velocity of an interferer refers to the displacement rate of an interferer in the image plane between consecutive image frames within the field of view. This velocity can be calculated using optical flow methods or feature point tracking algorithms, for example by analyzing the change in the interferer's pixel coordinates within the image sequence. This velocity is primarily used to infer the relative motion relationship between the interferer and the endoscope, thereby calculating its relative depth. The interference layer depth distribution map refers to an image or data structure that reflects the spatial depth distribution of the interferer within the field of view. This depth map can be represented as a depth map or a 3D point cloud, for example by mapping each interferer pixel to its corresponding depth value. This depth map primarily provides information about the interferer's position in 3D space, enabling effective removal or compensation during image reconstruction. The scanning operation refers to the rotation, swinging, or small-scale translation of the endoscope within the intestine. This can be achieved by performing a sector-shaped or spiral scan of the endoscope's tip within the intestine. This operation primarily captures images from different perspectives, accumulating clear areas not obscured by the interferer to construct a complete background image. The background canvas refers to a virtual image space used to accumulate and store clear image information that is not obscured by interference objects. It can be initialized with a blank image matrix or data structure, such as a two-dimensional image buffer corresponding to the field of view of the endoscope. Its main purpose is to gradually construct an interference-free, high-definition intestinal wall background image through the accumulation and fusion of multiple frames of images.

[0076] In some preferred embodiments, the present application is implemented as follows. The data quality guidance interface can be a real-time dashboard integrated into the endoscope console display, where the color information calibration completeness can be displayed as a percentage, for example, from 0% to 100%. When it reaches 95% or above, the indicator light turns green. The validity of the interference layer depth data can be displayed as textual statuses such as "good," "medium," and "poor," accompanied by corresponding color cues. Background area coverage can be presented as a real-time updated background canvas thumbnail and coverage percentage. In response to axial movement of the endoscope, the front end of the endoscope can be equipped with a distance sensor to accurately measure the distance between the endoscope and the intestinal wall, thereby automatically triggering the capture of narrowband light images at a first preset position (e.g., approximately 1 cm from the intestinal wall) and a second preset position (e.g., approximately 3 cm from the intestinal wall). An image processing unit, such as a high-performance embedded processor or an external computing server, receives these images. Brightness gain values ​​for the blue and green channels can be calculated by analyzing the average brightness change of specific regions in the image (e.g., a uniform region of the intestinal wall). The asymmetric scatter correction coefficient can be calculated based on the ratio of these gain values ​​or using a pretrained model. The coefficient's stability can be assessed by capturing multiple frames of imagery and calculating the standard deviation of the coefficient. When the standard deviation falls below a preset threshold, the coefficient is considered stable. In response to the endoscope's advancement, the image processing unit can utilize an optical flow algorithm or a deep learning-based motion estimation model to track the pixel movement velocity of interfering objects (such as suspended particles or mucus flocs) within the field of view. For example, by analyzing the displacement vectors of the interfering object's feature points in consecutive frames, its velocity on the image plane can be calculated. The relative depth of the interfering object can be calculated using triangulation or motion parallax-based depth estimation methods. For example, the known movement velocity of the endoscope and the pixel movement velocity of the interfering object can be used to infer its depth. The depth distribution map of the interfering layer can be a depth map with the same resolution as the image, storing a corresponding depth value for each pixel. The validity of the depth map can be determined by checking whether it contains sufficient valid depth points and whether the distribution of depth values ​​is reasonable. In response to the scanning operation of the endoscope, the image processing unit can analyze the captured image frames in real time and identify local areas that are not obscured by interference objects and have clarity that meets the requirements. The image information of these clear areas will be extracted and image registration will be performed to eliminate the displacement and rotation caused by the movement of the endoscope, and then accumulated into the background canvas. The background canvas can be a high-resolution image buffer that is iteratively updated to merge clear areas from different frames. For overlapping areas, higher quality image information can be selected for update based on the clarity evaluation results. The coverage of the background canvas can be updated in real time by calculating the ratio of the number of filled valid pixels to the total number of pixels. When all instructions are met, the image synthesis operation is triggered.The image processing unit first performs color compensation on the image information accumulated in the background canvas, applying the previously calculated asymmetric scatter correction coefficient to restore the image's true color. Subsequently, combined with the generated interference layer depth distribution map, the clear image in the background canvas is combined with the depth information through image fusion algorithms or 3D reconstruction techniques to reconstruct a complete, interference-free, clear image. Finally, the lesion detection module, such as an image recognition model based on a convolutional neural network (CNN), identifies and locates the lesion on this reconstructed clear image.

[0077] The present application further proposes that step S2 includes: responding to the axial movement of the endoscope, continuously capturing an image sequence for color correction during the axial movement; dividing each frame of the image sequence for color correction into a central area and an edge area; respectively calculating the brightness changes of the blue light channel and the green light channel in the central area, and the brightness changes of the blue light channel and the green light channel in the edge area; comparing the brightness changes of the blue light channel in the central area with the brightness changes of the blue light channel in the edge area, and comparing the brightness changes of the green light channel in the central area with the brightness changes of the green light channel in the edge area, to identify whether there is a particle exclusion effect in the central area The invention relates to a method for adjusting the brightness of the blue light channel in the edge area as follows: taking the brightness change of the blue light channel in the edge area as the real brightness change of the blue light channel reflecting the inherent optical attenuation characteristics under the body fluid interference environment, and taking the brightness change of the green light channel in the edge area as the real brightness change of the green light channel reflecting the inherent optical attenuation characteristics under the body fluid interference environment; calculating the brightness gain value of the blue light channel according to the real brightness change of the blue light channel; calculating the brightness gain value of the green light channel according to the real brightness change of the green light channel; calculating the asymmetric scattering correction coefficient according to the brightness gain values ​​of the blue light channel and the green light channel; and updating the indication of the completeness of the color information calibration according to the stability of the asymmetric scattering correction coefficient.

[0078] Among them, the particle exclusion effect refers to the local fluid dynamic disturbance caused by the front end of the endoscope on the suspended particles in the surrounding body fluids during the axial movement of the endoscope, which causes these particles to move rapidly or aggregate in a specific area of ​​the image field (such as the central area), thereby causing instantaneous and irregular changes in the image brightness or texture. Its purpose is to identify and distinguish the local interference caused by the movement of the endoscope from the attenuation characteristics of the medium itself; the inherent optical attenuation characteristics refer to the inherent attenuation law of brightness or energy caused by the absorption and scattering of the medium itself when light passes through the medium (including body fluids and tissues) in a body fluid interference environment. It can reflect the physical optical properties of the medium. Its purpose is to obtain stable light attenuation information that is not disturbed by instantaneous particle movement, providing a basis for accurate color correction.

[0079] In some preferred embodiments, the present application is implemented as follows: As the endoscope is moved axially within the intestine, such as by advancement or retraction, the system can continuously capture narrowband light images at a rate of 30 frames per second, forming an image sequence for color correction. Each frame in the image sequence can be divided into a central region and an edge region. For example, the central region can be defined as a rectangular area in the center of the image, occupying approximately 20% to 40% of the total image area, while the edge region can be defined as a peripheral annular region surrounding the central region. The system can then calculate the brightness changes in the blue and green channels of these two regions, respectively. The brightness change can be obtained by comparing the average pixel brightness values ​​of the corresponding region in the current frame with those in the previous frame, or by calculating the standard deviation of the average brightness of the region over a period of time. The system can then compare the brightness changes in the central and edge regions. For example, if the brightness change amplitude (such as the standard deviation or the maximum and minimum difference) in the central region is significantly greater than that in the edge region, it can be identified that a particle crowding effect is present in the central region. Once the particle crowding effect is identified, the system intelligently selects the brightness changes in the blue and green channels in the edge regions as the true brightness changes reflecting the inherent optical attenuation characteristics in the presence of body fluid interference. This is because the edge regions are less affected by the particle crowding effect, and their brightness changes better represent the true attenuation of light in body fluids. Based on these true brightness changes, the system calculates the brightness gain values ​​for the blue and green channels according to a preset optical model or lookup table. For example, the brightness change can be converted to optical path length based on the light attenuation model, thereby calculating the gain coefficient required to compensate for the light attenuation. Furthermore, based on the calculated brightness gain values ​​for the blue and green channels, the system calculates the asymmetric scatter correction coefficient. This coefficient, which can be a matrix or a set of parameters, is used in subsequent image processing to differentially compensate for different color channels to correct for color distortion caused by body fluid scattering. Finally, the system continuously monitors the numerical fluctuations of the asymmetric scatter correction coefficient. If the coefficient remains within a preset stable range in multiple consecutive frames of images, the indication of the color information calibration completeness can be updated to "achieved" or "high completeness", otherwise it may be indicated as "pending correction" or "low completeness", thereby providing real-time quality feedback to the operator.

[0080] The present application further proposes that the steps of tracking the pixel movement speed of the interference object in the field of view during the pushing operation include: responding to the pushing operation of the endoscope, continuously capturing the image sequence used for depth estimation during the pushing operation; identifying the image area where the interference object is located in the field of view from the image sequence used for depth estimation; for the image area where the interference object is located, estimating the motion vector of the pixel point in the image area where the interference object is located between consecutive frames of the image sequence used for depth estimation; and calculating the pixel movement speed of the interference object based on the motion vector of the pixel point.

[0081] Identifying the image region where the distractor is located within the field of view refers to determining which pixels or regions in the image belong to the distractor. This can be achieved through image segmentation techniques, such as threshold segmentation based on color, texture, or brightness differences, or by using deep learning models, such as convolutional neural networks, to perform semantic segmentation on the image, directly identifying the distractor category and locating its region. This is done to reduce the calculation scope of subsequent motion estimation and improve processing efficiency and accuracy. Furthermore, estimating the motion vector of the pixel points within the image region where the distractor is located refers to quantifying the displacement and direction of the pixel points in the image between consecutive frames. Specifically, optical flow methods, such as the Lucas-Kanade sparse optical flow or the Farneback dense optical flow algorithm, can be used to calculate the motion vector of each pixel or specific feature point in the image. This aims to capture the fine motion information of the distractor in the field of view and provide accurate data for subsequent velocity calculations.

[0082] In some preferred embodiments, in response to the endoscope's advancement, the system can continuously capture image sequences for depth estimation at a rate of 30 frames per second, ensuring the acquisition of sufficiently dense time-series image data. Specifically, from these continuously captured image sequences, the image regions containing interfering objects within the field of view can be identified using a background difference-based method, namely, by comparing the pixel differences between the current frame and a preset, non-interfering background frame, or by using a pre-trained convolutional neural network model to perform real-time semantic segmentation of the image, thereby accurately identifying and delineating the image regions containing suspended particles or mucus flocs in the liquid. Furthermore, for these identified interfering image regions, the Farneback dense optical flow algorithm can be used between consecutive frames of the image sequence used for depth estimation to estimate the motion vector of each pixel within the image region containing the interfering object. This algorithm can calculate the displacement vector of each pixel in the image, thereby comprehensively reflecting the motion trajectory and deformation of the interfering object. Ultimately, based on the pixel motion vectors obtained from these dense optical flow fields, the pixel movement speed of the interfering object on the image plane can be calculated. For example, by calculating the modulus of the motion vector and normalizing it with the frame rate, quantitative velocity information can be obtained.

[0083] The present application further proposes performing a composite detection action in response to a propulsion operation of an endoscope, wherein the composite detection action includes a uniform propulsion phase and a momentary pause phase, and includes the following steps: in the momentary pause phase, classifying the particles in the interference object into first inertial particles and second inertial particles based on the motion velocity attenuation characteristics of the particles in the interference object; in the uniform propulsion phase, calculating the relative depth of the first inertial particles based on the pixel movement velocity of the first inertial particles; determining the distance to the intestinal wall mucosa based on the relative depth of the first inertial particles; determining the spatial area between the front end of the endoscope and the intestinal wall mucosa as the fluid layer depth range based on the distance to the intestinal wall mucosa; setting the relative depth of the second inertial particles to a preset depth or depth range within the fluid layer depth range; integrating the relative depth information of the first inertial particles with the relative depth information of the second inertial particles to generate an interference layer depth distribution map; and updating an indication of the validity of the interference layer depth data based on the validity of the interference layer depth distribution map.

[0084] Among them, the composite detection action refers to the detection behavior that combines different motion modes during the advancement of the endoscope. It can be achieved by a pre-programmed motion sequence or a specific rhythm manually controlled by the operator. Its purpose is to provide observation conditions for distinguishing the motion characteristics of different types of interfering particles; the uniform propulsion stage refers to the period when the endoscope moves forward at a constant speed. It can be achieved by stepper motor control or steady propulsion by the operator. Its purpose is to observe the pixel movement speed of particles in a stable motion state; the instantaneous pause stage refers to the period when the endoscope stops moving briefly during the advancement process. It can be achieved by a braking mechanism or the operator instantly stopping the advancement. Its purpose is to classify the particles by using the inertia difference; the motion speed attenuation characteristic refers to the law that the particle's own motion speed gradually decreases with time after the endoscope stops moving. It can be quantified by analyzing the displacement change rate of the particles in continuous frame images or tracking the particle trajectory through the optical flow algorithm. Its purpose is to distinguish the inertial behavior of particles of different densities or viscosity; First inertial particles refer to particles whose motion speed decays slowly during the instantaneous pause phase. They can be solid particles with higher density or other impurities with higher inertia. Their purpose is to serve as a reliable reference point for depth estimation. Second inertial particles refer to particles whose motion speed decays faster during the instantaneous pause phase. They can be mucus flocs with higher viscosity or suspended solids with lower density. Their purpose is to distinguish them from first inertial particles and adopt different depth estimation strategies. The fluid layer depth range refers to the liquid-filled spatial region between the endoscope tip and the intestinal wall mucosa. It can be determined based on the focal length information of the endoscope or through optical triangulation methods. Its purpose is to provide reasonable constraints for the depth estimation of the second inertial particles. The preset depth or depth range refers to a fixed depth value or an allowable depth interval set for the second inertial particle. It can be determined based on empirical data, fluid dynamics models, or real-time environmental parameters. Its purpose is to provide a reasonable estimate when the depth of the second inertial particle is difficult to accurately measure.

[0085] The solution of this application effectively improves the accuracy of the interference layer depth distribution map by introducing a refined interference depth estimation strategy. Specifically, when responding to the endoscope's propulsion operation, the system no longer simply tracks the pixel movement speed of all interference objects, but actively performs a composite detection action that cleverly combines a uniform propulsion phase and a momentary pause phase. During the uniform propulsion phase, the endoscope moves forward at a steady speed. At this time, the pixel movement speed of the first inertial particles in the field of view, such as dense solid residue, is clearly correlated with the propulsion speed of the endoscope. Based on the pixel movement speed of these particles, the system can use the principle of motion parallax to more accurately calculate their relative depth relative to the front end of the endoscope. More importantly, during the momentary pause phase, the endoscope will briefly stop moving. At this time, the system will classify the different particles in the interference objects into first inertial particles and second inertial particles based on their motion speed attenuation characteristics. For example, due to inertia, the motion speed of dense solid particles decays more slowly, while the more viscous mucus flocs will stop moving quickly or exhibit different attenuation patterns. This classification mechanism enables the system to identify and distinguish different types of interference objects. After obtaining the relative depth of the first inertial particle, the system can further use this depth information to infer the distance to the intestinal mucosa. This is because the first inertial particle is typically located in the fluid layer, and its depth estimation is relatively reliable, serving as a reference for determining the location of the intestinal mucosa. Once the distance to the intestinal mucosa is determined, the spatial region between the endoscope tip and the intestinal mucosa is defined as the fluid layer depth range. For interference objects classified as second inertial particles during the momentary pause, their complex motion characteristics make it difficult to accurately estimate their depth using simple pixel movement speed. Therefore, this solution sets their relative depth to a preset depth or depth range within the fluid layer depth range. This processing method avoids errors introduced by inaccurate estimation while ensuring that the depth information of the second inertial particle is within a reasonable range. Finally, the system integrates the relative depth information of the first and second inertial particles to generate a more comprehensive and accurate interference layer depth distribution map. This distribution map not only contains the precise depth of solid particles, but also reasonably includes the depth information of mucus flocs. Based on this more effective interference layer depth distribution map, the system can update the indication of the validity of the interference layer depth data, providing more reliable depth data for subsequent image reconstruction, improving the quality of the final reconstructed image and the accuracy of lesion detection. This multi-stage, multi-type particle differentiation depth estimation method overcomes the problem of insufficient accuracy of a single depth estimation model in complex interference environments, enabling endoscopic image reconstruction to more effectively remove interference and present clear mucosal details.

[0086] In some preferred embodiments, the present application is implemented as follows. While the endoscope is advancing, the control system can issue instructions causing the endoscope to perform a composite detection operation. For example, the endoscope can be advanced at a constant speed of 5 mm per second for approximately 2 seconds, followed by a momentary pause of 0.5 seconds. During the constant-speed advancement phase, the image processing module can continuously capture a sequence of images and track the pixel movement speed of all visible particles within the field of view using optical flow or a feature point matching algorithm. Particles that exhibit a stable pixel movement speed consistent with the endoscope's advancement direction during the constant-speed advancement phase—for example, if their pixel movement speed remains within a specific threshold range in each image frame and their motion trajectory is highly correlated with the axial movement direction of the endoscope—can be preliminarily identified as potential first-inertial particles. When the endoscope enters the momentary pause phase, the image processing module continues to capture images and analyzes the velocity decay characteristics of these particles. Specifically, the rate at which the particle pixel movement speed decreases in the first few frames after the pause can be monitored. For example, if a particle's pixel velocity decreases by more than 90% within 0.1 seconds after a pause, it can be classified as a second-inertial particle, typically corresponding to mucus flocs. If its pixel velocity decreases more slowly, for example, by less than 30% within 0.1 seconds, it can be classified as a first-inertial particle, typically corresponding to dense solid debris. During the uniform propulsion phase, for objects classified as first-inertial particles, the system can calculate the relative depth of these first-inertial particles based on their pixel velocity, the known propulsion speed of the endoscope, and camera parameters, using triangulation or a motion parallax model. For example, if a first-inertial particle moves X pixels per image frame, and the endoscope moves Y millimeters per frame, its distance from the endoscope tip can be calculated based on the pre-calibrated camera focal length and pixel size. The system can then use this relative depth information from these first-inertial particles to determine the distance to the intestinal mucosa. For example, the average depth of the group of first inertial particles closest to the endoscope can be selected as the reference distance to the intestinal mucosa, or the depth distribution of the first inertial particles can be analyzed to identify the farthest and most stable depth value as the distance to the intestinal mucosa. Once the distance to the intestinal mucosa is determined, the spatial region between the endoscope tip and the intestinal mucosa is defined as the fluid layer depth range. For example, if the intestinal mucosa is 15 mm from the endoscope tip, the fluid layer depth range can be set to the area from the endoscope tip to 15 mm. For objects classified as second inertial particles, due to their complex motion characteristics and difficulty in accurate measurement, the system can set their relative depth to a preset depth or depth range within the fluid layer depth range. For example, the depth of all second inertial particles can be uniformly set to the center value of the fluid layer depth range, or to a smaller depth range, such as between the first 20% and 50% of the fluid layer depth range.Finally, the system integrates the calculated relative depth information of the first inertial particles with the relative depth information of the set second inertial particles to generate a interference layer depth distribution map containing depth information for different types of interference objects. For example, a three-dimensional depth map can be created in which the depth of the first inertial particles is the exact calculated value, while the depth of the second inertial particles is a preset value or range. Based on the completeness and consistency of this interference layer depth distribution map, the system can update the indication of the validity of the interference layer depth data. For example, if the distribution map covers most of the interference objects in the field of view and the depth information has no obvious anomalies, the indication will be "valid."

[0087] The present application further proposes that step S4 includes: responding to the scanning operation of the endoscope, continuously capturing the image sequence used for background area accumulation in the scanning operation; for each frame of the image in the image sequence used for background area accumulation, identifying the local area in the frame of the image that meets the clarity requirements to obtain a clear area; evaluating the optical uniformity of each clear area to quantify its residual interference level and generate a corresponding clarity quality parameter; storing the image information of each clear area and its clarity quality parameter together; performing an image registration operation on the image information of the clear area, and accumulating the registered image information to the background canvas; in the process of accumulating the registered image information, for the image information corresponding to the clear area position stored in the background canvas, comparing the registered image information of the clear area with the image information stored at the corresponding position, and based on the numerical values ​​of their respective clarity quality parameters, selecting the image information with a higher clarity quality parameter value for updating or fusion into the background canvas; calculating the coverage of the background canvas based on the ratio of the number of valid pixels filled in the background canvas to the total number of pixels, and updating the indication of the background area information coverage.

[0088] The definition of clarity refers to the threshold or standard set for the clarity of a local area of ​​an image. Specifically, it can be quantified by calculating the image gradient, edge sharpness, contrast, or high-frequency components of the Fourier transform. The purpose is to screen out parts of the image with higher visual quality. The definition of a clear area refers to a local image area that meets the preset clarity requirement. Specifically, it can be obtained by performing a local clarity assessment on the image frame and selecting areas with clarity scores above a specific threshold. The purpose is to ensure that the image information processed subsequently has a high initial quality. The definition of optical uniformity refers to the degree of consistency in the distribution of illumination, color, and texture in the image area. Specifically, it can be assessed by analyzing the brightness, color, and texture fluctuations of pixels within the image area. The purpose is to quantify the extent to which the image area is affected by residual interference such as body fluids and bubbles. The degree of residual interference refers to the degree of optical distortion or information loss caused by the presence of non-target substances such as body fluids, bubbles, and mucus in the image area. Specifically, it can be determined by quantitatively analyzing the optical uniformity of the image area. The purpose is to identify and quantify minor interferences that still exist in the image and affect image quality. In [1], the clarity quality parameter refers to a numerical indicator used to comprehensively measure the clarity and optical uniformity of an image region. Specifically, it can be generated by combining the clarity assessment results and the quantitative value of the residual interference degree. Its purpose is to provide an objective quality basis for the selection and fusion of image information. The image registration operation refers to the process of geometrically aligning images at different times or different perspectives. Specifically, it can be achieved through algorithms such as feature point matching, optical flow, or direct pixel intensity optimization. Its purpose is to eliminate image displacement and deformation caused by endoscope movement and ensure that corresponding areas of different image frames can be accurately superimposed. The background canvas refers to the virtual image space used to accumulate and store image information obtained during endoscopic scanning that is not obscured by interference objects. Specifically, it can be a dynamically updated two-dimensional pixel matrix. Its purpose is to construct a complete and clear panorama of the intestinal mucosa. The coverage refers to the ratio of the number of valid pixels filled in the background canvas to the total number of pixels. Specifically, it can be calculated by counting the number of non-empty pixels in the background canvas and dividing it by the total number of pixels. Its purpose is to indicate the completeness of the background canvas and provide a completion reference for subsequent image synthesis operations.

[0089] In some preferred embodiments, the present application is implemented as follows. While the endoscope is scanning the intestine, the image acquisition module can continuously capture a 30-frame-per-second video stream as an image sequence for background region accumulation. The image processing unit can execute a clarity assessment algorithm on each image frame, such as using the Laplacian operator or the Tenengrad gradient function, to calculate a local clarity score for the image. If the clarity score of a local region exceeds a preset threshold, such as 0.8 (after normalization), the region is identified as a clear region. For each identified clear region, the image processing unit can further assess its optical uniformity. This can include dividing the clear region into multiple subregions and performing a fluctuation analysis on the brightness, color (e.g., mean and variance of RGB channels), and texture (e.g., local binary pattern (LBP) features) of each subregion across consecutive frames. For example, the standard deviation of pixel brightness in the temporal dimension, the mean squared deviation of the color channels, and the entropy of the texture features can be calculated for each subregion. These fluctuation amplitudes can be weighted and fused to generate a residual interference measure for that subregion. The residual interference measures of all subregions can be combined to quantify the residual interference level of the entire clear region. The sharpness quality parameter can be composed of a weighted average of the sharpness score and the residual noise level, for example, the sharpness score multiplied by 0.7 plus (1-residual noise level) multiplied by 0.3. The image information of these sharp areas and their corresponding sharpness quality parameters can be stored in memory or cache. The image processing unit can then perform image registration on these sharp areas. This can employ a feature point matching method, such as SIFT or ORB features, combined with the RANSAC algorithm to estimate the homography matrix between the images, thereby accurately aligning the sharp areas from different frames to the coordinate system of the background canvas. When accumulating the registered image information into the background canvas, the image processing unit can examine each pixel in the background canvas. If a pixel in the current registered sharp area overlaps with an existing pixel in the background canvas, the system can compare the sharpness quality parameters of the two pixels. For example, if the quality parameter of the current pixel is 0.9 and the quality parameter of the corresponding pixel in the background canvas is 0.7, the system will select the current pixel with the higher quality parameter to update the corresponding pixel in the background canvas. If there is no valid pixel information at the corresponding position in the background canvas, the current pixel information is directly filled in. This selective accumulation strategy ensures that the background canvas is always composed of image information of the best quality. Finally, the system can monitor the filling status of the background canvas in real time. By counting the number of non-zero pixels in the background canvas and comparing it with the total number of pixels in the background canvas, the coverage of the background canvas can be calculated. For example, if the background canvas is a 1024x768 pixel image, of which 500,000 pixels have been effectively filled, the coverage is This coverage value can be displayed on the user interface as an indication of the coverage of the background area information, informing the operator of the completeness of the background canvas.

[0090] The present application further proposes the steps of evaluating the optical uniformity of each clear area to quantify its residual interference level, including: dividing the clear area into multiple sub-areas; capturing an image sequence corresponding to the clear area; for each sub-area in the image sequence corresponding to the clear area, calculating the brightness change, color change, and texture change of the pixels in each sub-area to evaluate the optical uniformity of each sub-area; generating a corresponding residual interference metric based on the brightness change, color change, and texture change of the pixels in each sub-area; and comprehensively quantifying the residual interference level of the clear area based on the residual interference metrics of all sub-areas.

[0091] Among them, a clear area refers to a local image area that is identified in the image as meeting certain clarity requirements, which may contain mucosal tissue information to be observed; a sub-area refers to a smaller, independent image unit obtained by further subdividing a larger clear area, which can be achieved by grid division, adaptive segmentation or clustering based on content features, etc., and its purpose is to perform a fine local analysis of the optical properties inside the clear area; an image sequence refers to a series of image frames captured continuously over a period of time, which can be composed of a video stream or image frames continuously taken by a high-speed camera, and its purpose is to capture the optical properties of the clear area in the time dimension; brightness change refers to the difference in light intensity values ​​of image pixels or areas at different time points or spatial positions, which can be manifested as fluctuations in pixel grayscale values ​​or shifts in average brightness, and its purpose is to reflect the stability and uniformity of lighting conditions; color change refers to the difference in light intensity values ​​of image pixels or areas at different time points or spatial positions. The difference in color information in spatial position can be manifested as fluctuations in the values ​​of color space components such as RGB or HSV, and its purpose is to reflect the fidelity and consistency of color; texture change refers to the difference in local structure or detail pattern of image pixels or regions at different time points or spatial positions, which can be manifested as fluctuations in local contrast, edge density or frequency characteristics, and its purpose is to reflect the clarity of image details and the presence of interference; residual interference measure refers to a quantitative indicator of the degree of deviation from optical uniformity in a sub-region, which can be a comprehensive value obtained by weighted or unweighted combination of brightness, color and texture changes, and its purpose is to convert the interference degree of the sub-region into a comparable value; residual interference degree refers to an overall quantitative evaluation of residual interference in the entire clear area, which can be obtained by statistical analysis or aggregate calculation of the residual interference measures of all sub-regions, and its purpose is to provide an overall quality assessment of the clear area.

[0092] In some preferred embodiments, the present application is implemented as follows: When the system identifies a local region within an image frame that meets clarity requirements, i.e., a clear region, to perform a detailed assessment of its optical uniformity, the system first divides the clear region into a fixed-size grid, for example, into multiple subregions of 8x8 or 16x16 pixels. The system then captures a continuous image sequence corresponding to the clear region over a period of time, for example, by extracting 30 frames of images of the clear region within the last 1 second from an endoscope video stream. For each subregion in the image sequence, the system calculates the brightness variation of the pixels within the subregion, for example, by calculating the standard deviation of the brightness values ​​of all pixels in the subregion at the corresponding position in the image sequence to reflect brightness fluctuations; simultaneously, the system calculates its color variation, for example, by calculating the Euclidean distance of the RGB color components of all pixels in the subregion at the corresponding position in the image sequence to reflect color fluctuations; and finally, calculates its texture variation, for example, by calculating the variance of the local gradient amplitude of all pixels in the subregion at the corresponding position in the image sequence to reflect texture fluctuations. These calculation results are used to evaluate the optical uniformity of each subregion. Next, a corresponding residual interference metric is generated based on the brightness, color, and texture changes of the pixels within each subregion. For example, these three change values ​​can be summed or weighted to obtain a numerical value representing the interference level of the subregion. Finally, based on the residual interference metrics of all subregions, for example, the average or maximum value of the residual interference metrics of all subregions can be calculated to comprehensively quantify the residual interference level of the entire clear region, resulting in a clarity quality parameter representing the overall quality of the clear region.

[0093] The present application further proposes that for each sub-region in the image sequence corresponding to the clear region, the steps of calculating the brightness change, color change and texture change of pixels in each sub-region include: obtaining the brightness information of the pixels in the sub-region in the time dimension, and calculating the brightness fluctuation amplitude of the pixels in the sub-region based on the brightness information in the time dimension; obtaining the color information of the pixels in the sub-region in the time dimension, and calculating the color fluctuation amplitude of the pixels in the sub-region based on the color information in the time dimension; obtaining the texture information of the pixels in the sub-region in the time dimension, and calculating the texture fluctuation amplitude of the pixels in the sub-region based on the texture information in the time dimension.

[0094] Among them, obtaining the brightness information of the pixels in the sub-region in the time dimension means that during the endoscopic scanning process, for each sub-region in the clear area, the brightness value sequence corresponding to each pixel point in the sub-region is continuously collected at continuous time points. This can be achieved by extracting the grayscale value or brightness channel value of the corresponding pixel point from the continuously captured image sequence, and its purpose is to provide time series data for the subsequent brightness fluctuation analysis; calculating the brightness fluctuation amplitude of the pixels in the sub-region means quantifying the degree of change of the brightness of the pixels in the sub-region over time based on the acquired brightness value sequence, and it can be achieved by calculating the standard deviation, root mean square error or maximum and minimum value difference of the brightness sequence, and its purpose is to characterize the degree of brightness instability caused by interference; obtaining the color information of the pixels in the sub-region in the time dimension means that during the endoscopic scanning process, for each sub-region in the clear area, the color value sequence corresponding to each pixel point in the sub-region is continuously collected at continuous time points. This can be achieved by extracting the RGB, HSV or Lab color space component values ​​of the corresponding pixel point from the continuously captured image sequence, and its purpose is to provide time series data for the subsequent color fluctuation analysis; calculating the color of the pixels in the sub-region Color fluctuation amplitude refers to quantifying the degree of temporal change of pixel color in a sub-region based on the acquired color value sequence. This can be achieved by calculating statistics such as the standard deviation of the color component sequence, the average change of color space distance, or the dynamic change of the color histogram. Its purpose is to characterize the degree of color distortion or drift caused by interference. Acquiring the texture information of pixels in the sub-region in the time dimension refers to continuously collecting the local texture feature sequence corresponding to each pixel point in the sub-region at continuous time points for each sub-region in the clear area during the endoscopic scanning process. This can be achieved by extracting the gradient information of the corresponding pixel point, local binary pattern (LBP) feature, or Gabor filter response from the continuously captured image sequence. Its purpose is to provide time series data for subsequent texture fluctuation analysis. Calculating the texture fluctuation amplitude of pixels in the sub-region refers to quantifying the degree of temporal change of pixel texture in the sub-region based on the acquired texture feature sequence. This can be achieved by calculating statistics such as the standard deviation of the texture feature value sequence, the average change of the texture feature vector distance, or the dynamic change of texture energy. Its purpose is to characterize the degree of blurring or distortion of image details caused by interference.

[0095] In some preferred embodiments, for each subregion in the image sequence corresponding to the clear region, the step of calculating the brightness, color, and texture changes of the pixels within each subregion can be specifically implemented as follows: the system can first obtain the temporal brightness information of the pixels within the subregion. This can involve extracting the brightness values ​​of each pixel within each subregion at different time frames (e.g., N consecutive image frames) from a continuously captured image sequence. For example, for a pixel (x, y), a sequence of brightness values ​​L(x, y, t1), L(x, y, t2), ..., L(x, y, tN) can be obtained. Subsequently, based on this temporal brightness information, the brightness fluctuation amplitude of the pixels within the subregion can be calculated. One calculation method is to calculate the average standard deviation of the brightness sequence for all pixels within the subregion, or to calculate the peak-to-valley difference (maximum value minus minimum value) of the brightness sequence for each pixel. The average or maximum value of these peak-to-valley differences is then taken as the brightness fluctuation amplitude for the subregion. Next, the system can obtain the temporal color information of the pixels within the subregion. This can involve extracting the color component values ​​of each pixel across N consecutive image frames, for example, a sequence of R, G, and B component values ​​in the RGB color space. For a pixel (x, y), one can obtain R(x, y, t1)...R(x, y, tN), G(x, y, t1)...G(y, y, tN), B(x, y, t1)...B(x, y, tN), and so on. Then, based on this temporal color information, the color fluctuation amplitude of pixels within the sub-region can be calculated. This can include calculating the standard deviation of the R, G, and B component sequences separately and taking a weighted sum of them, or calculating the Euclidean distances between pixel color vectors in adjacent frames in color space and performing statistical analysis on these distances (e.g., calculating the average distance or maximum distance) to characterize color fluctuations. Finally, the system can obtain temporal texture information for pixels within the sub-region. This can involve extracting local texture features from N consecutive frames of images for each pixel point and its neighborhood within the sub-region, for example, by calculating the mean or variance of the image gradient amplitude, or applying the local binary pattern (LBP) operator to extract texture descriptors. Then, based on the texture information in these time dimensions, the texture fluctuation amplitude of the pixels in the sub-region can be calculated. This can include calculating the standard deviation of the texture feature value sequence, or calculating the similarity change of the texture feature vectors between consecutive frames, for example, by calculating the cosine similarity of the texture feature vectors and quantifying the degree of their change over time. Through these specific calculation methods, accurate brightness, color and texture fluctuation amplitudes can be obtained, thereby providing a reliable quantitative basis for subsequent optical uniformity evaluation.

[0096] The present application further proposes the steps of generating corresponding residual interference measurements based on the brightness changes, color changes and texture changes of pixels in each sub-region, including: performing weighted fusion based on the brightness fluctuation amplitude, color fluctuation amplitude and texture fluctuation amplitude of pixels in the sub-region to obtain the residual interference measurement of the sub-region.

[0097] Among them, weighted fusion refers to the process of assigning different weights to multiple input quantities according to their importance or contribution, and then combining the weighted input quantities to obtain a comprehensive output value. It can be implemented by linear weighting, nonlinear weighting or fusion based on machine learning models. Its purpose is to comprehensively consider the impact of information of different dimensions on the final result; residual interference metric refers to a numerical indicator that quantifies the degree of image quality degradation in an image sub-region due to various interference factors. It can be a normalized value, such as a range from 0 to 1, where 0 indicates no interference and 1 indicates severe interference. Its purpose is to provide a quantitative basis for subsequent image quality evaluation and background canvas construction.

[0098] In some preferred embodiments, the present application is specifically implemented as follows. When performing weighted fusion based on the brightness fluctuation amplitude, color fluctuation amplitude, and texture fluctuation amplitude of pixels in a sub-region, a weight determination method based on expert experience or machine learning training can be adopted. For example, a set of weights can be preset, such as the brightness fluctuation amplitude weight is W_L, the color fluctuation amplitude weight is W_C, and the texture fluctuation amplitude weight is W_T, where the sum of these weights can be 1. The residual interference measure of the sub-region can be calculated by the following formula: Residual interference measure (Brightness fluctuation range) (Color fluctuation range) (Texture Fluctuation Amplitude). These weights can be adjusted based on the types of interference commonly found in endoscopic images and their impact on image features. For example, if a certain interference (such as an air bubble) is known to primarily cause large variations in brightness while having relatively little impact on color and texture, the weight for brightness fluctuation can be increased.

[0099] As a specific implementation, dynamic adjustment rules can also be used to determine the weights. For example, the system can adjust the weights of each fluctuation amplitude in real time based on the overall characteristics of the current image sequence or the main interference type detected. If a large area of ​​mucus coverage is detected in the image, the system can increase the weight of the color fluctuation amplitude because mucus usually causes color distortion. If fast-moving particles are detected, the weights of the brightness fluctuation amplitude and texture fluctuation amplitude can be increased. This dynamic adjustment can be achieved through a pre-trained classifier or adaptive algorithm, which can automatically output the corresponding weight combination based on the input image features. In this way, the residual interference metric can be made adaptive and able to cope with endoscopic inspection environments of different complexities.

[0100] The present application further proposes to perform weighted fusion according to the brightness fluctuation amplitude, color fluctuation amplitude and texture fluctuation amplitude of pixels in the sub-region to obtain the residual interference measurement of the sub-region, including the following steps: determining the weights corresponding to the brightness fluctuation amplitude, color fluctuation amplitude and texture fluctuation amplitude of pixels in the sub-region based on preset rules or dynamic adjustment rules; performing weighted fusion according to the brightness fluctuation amplitude, color fluctuation amplitude, texture fluctuation amplitude and corresponding weights of pixels in the sub-region to obtain the residual interference measurement of the sub-region.

[0101] Preset rules refer to rules pre-set before the system runs based on experience, experimental data, or domain knowledge to determine the weights associated with each fluctuation amplitude. These rules can be implemented using fixed ratios, lookup tables, or rule sets based on specific scene classifications. Dynamic adjustment rules refer to rules that automatically adjust the weights associated with each fluctuation amplitude during system operation based on real-time image data, environmental information, or algorithmic analysis results. These rules can be implemented using machine learning-based models, adaptive algorithms, or feedback control mechanisms. Weights refer to the influence coefficients or importance factors assigned to the brightness, color, and texture fluctuation amplitudes during weighted fusion. These weights can be numerical, such as floating-point numbers between 0 and 1, or integer ratios. Weighted fusion involves combining multiple values ​​of varying importance or influence (here, brightness, color, and texture fluctuation amplitudes) according to their corresponding weights to produce a comprehensive metric (here, the residual interference metric). This can be implemented using weighted averaging, weighted summation, or other linear or nonlinear combinations.

[0102] The solution of this application calculates the residual interference metric for a subregion by weighted fusion of the brightness, color, and texture fluctuation amplitudes of pixels within the subregion. Specifically, the system first determines the corresponding weights for the brightness, color, and texture fluctuation amplitudes based on preset or dynamically adjusted rules. This allows the system to flexibly adjust the contribution of these fluctuation amplitudes to the residual interference metric based on different imaging environments or interference types. For example, when the image is primarily affected by uneven illumination, the weight of the brightness fluctuation amplitude can be increased; when the image is primarily affected by stray light, the weight of the color fluctuation amplitude can be increased; and when the image is primarily affected by fine particle interference, the weight of the texture fluctuation amplitude can be increased. This weighting mechanism enables the residual interference metric to more accurately reflect the primary interference type and degree within the current image. Based on this, the system performs a weighted fusion calculation based on the determined brightness, color, and texture fluctuation amplitudes and their corresponding weights to obtain the residual interference metric for the subregion. Through this weighted fusion, the impact of different types of interference factors on image quality is comprehensively considered, and their impact is accurately reflected through the weights. Interference factors with higher weights have a greater impact on the final residual interference measurement due to their fluctuation amplitude; conversely, interference factors with lower weights have a smaller impact on the final measurement due to their fluctuation amplitude. This refined residual interference measurement method is closely integrated with the overall process of evaluating the optical uniformity of clear areas in this application. By more accurately quantifying the degree of residual interference in each sub-area, more reliable clarity quality parameters can be generated. These clarity quality parameters play a key role in the background canvas construction process, guiding the system to prioritize and fuse image information with lower residual interference and better optical uniformity during image registration and accumulation. Therefore, this solution not only improves the accuracy of the residual interference measurement of a single sub-area, but also further optimizes the construction quality of the background canvas, so that the final reconstructed clear image can more effectively remove interference, providing a purer and more reliable visual basis for subsequent lesion detection.

[0103] In some preferred embodiments, the present application is implemented as follows: a hybrid strategy can be employed to determine the weights corresponding to the brightness, color, and texture fluctuation amplitudes of pixels within a subregion. For example, several sets of weight configurations can be preset, each corresponding to a typical interference scenario, such as "primarily turbid fluid," "primarily particulate matter interference," or "combined interference." When the system detects that the overall characteristics of the current image sequence (for example, by analyzing the image's average brightness, color saturation, or edge density) match a preset scenario, the preset weights corresponding to that scenario can be loaded. As a specific implementation, dynamic adjustment rules can also be employed. For example, the system can analyze the statistical characteristics of pixels within a subregion in real time. If the brightness fluctuation amplitude is significantly higher than the color and texture fluctuation amplitudes, the weight of the brightness fluctuation amplitude can be dynamically increased, while the weights of the color and texture fluctuation amplitudes can be appropriately decreased. The reverse is also true. This dynamic adjustment can be based on a simple heuristic algorithm, for example, setting the weights as functions proportional to the corresponding fluctuation amplitudes, or using a small neural network model to output the optimal weight combination based on the input brightness, color, and texture fluctuation amplitudes, as well as other contextual information (such as the distance between the endoscope and the intestinal wall and the status of the irrigation fluid injection). Once these weights are determined, for example, the brightness fluctuation amplitude weight W_L, the color fluctuation amplitude weight W_C, and the texture fluctuation amplitude weight W_T, the system can perform weighted fusion based on the brightness fluctuation amplitude L_amp, color fluctuation amplitude C_amp, and texture fluctuation amplitude T_amp of the pixels in the sub-region. Specifically, the residual interference measure R_D of the sub-region can be calculated as: In this way, a comprehensive and accurate value reflecting the residual interference level of the sub-region can be obtained.

[0104] refer to Figure 2The present application further proposes a fiber endoscope image lesion detection system, which is applied to a fiber endoscope image lesion detection method. The system includes: a data quality guidance module for providing a data quality guidance interface and indicating the completeness of color information calibration, the validity of interference layer depth data and the coverage of background area information; a color correction module for responding to the axial movement of the endoscope, capturing the narrowband light images of the endoscope at the first preset position and the second preset position away from the intestinal wall during the axial movement, calculating the brightness gain values ​​of the blue light channel and the green light channel based on the narrowband light images at the first preset position and the second preset position, calculating the asymmetric scattering correction coefficient based on the brightness gain values ​​of the blue light channel and the green light channel, and updating the indication of the color information calibration completeness based on the stability of the asymmetric scattering correction coefficient; a depth estimation module for responding to the advancement operation of the endoscope, tracking the pixel movement of interference objects in the field of view during the advancement operation. The moving speed of the interference object is calculated according to the pixel moving speed of the interference object, and the relative depth of the interference object is obtained to obtain the interference layer depth distribution map, and according to the validity of the interference layer depth distribution map, the indication of the validity of the interference layer depth data is updated; the background construction module is used to respond to the scanning operation of the endoscope, identify the image area not blocked by the interference object, store the image information of the image area not blocked by the interference object to the background canvas, and update the indication of the background area information coverage according to the coverage of the background canvas; the image synthesis module is used to trigger the image synthesis operation when the color information calibration completeness, the interference layer depth data validity and the background area information coverage indication have been achieved, and perform color compensation on the image information of the image area not blocked by the interference object stored in the background canvas according to the asymmetric scattering correction coefficient, and reconstruct a clear image in combination with the interference layer depth distribution map and the background canvas; the lesion detection module is used to perform lesion detection on the clear image.

[0105] Among them, the data quality guidance module refers to a logic unit for providing a user interaction interface and data status feedback, which can be a software interface component, the purpose of which is to show the operator the integrity and effectiveness of the current data acquisition and processing in real time, so as to guide the operation; the color correction module refers to a processing unit for adjusting the image color to eliminate optical distortion, which can be a set of image processing algorithms, the purpose of which is to correct the asymmetric color deviation caused by liquid scattering; the depth estimation module refers to a processing unit for calculating the relative distance of objects in the field of view, which can be an algorithm based on visual odometry or optical flow analysis, the purpose of which is to obtain the spatial depth information of the interference object; the background construction .... A block refers to a logical unit used to accumulate and integrate image information of interference-free areas to form a complete background image. Specifically, it can be an image stitching and fusion algorithm, the purpose of which is to gradually establish a clear background image of the intestinal wall; an image synthesis module refers to an integrated unit used to fuse image information at different processing stages to generate a final clear image. Specifically, it can be an image rendering engine, the purpose of which is to combine the color-compensated background image with depth information to reconstruct high-quality visual content; a lesion detection module refers to an analysis unit used to identify and mark abnormal areas on the processed image. Specifically, it can be a diagnostic algorithm based on deep learning or traditional image analysis, the purpose of which is to assist doctors in detecting potential lesions.

[0106] The contents disclosed above are only preferred feasible embodiments of the present invention and do not limit the scope of protection of the present invention. Therefore, all equivalent technical changes made using the contents of the present invention description and drawings are included in the scope of protection of the present invention. In addition, the elements therein can be updated as technology develops.

Claims

1. A fiber endoscope image lesion detection method, characterized in that: The method comprises the following steps: S1: Provides a data quality guidance interface, which is used to indicate the completeness of color information calibration, the validity of interference layer depth data, and the coverage of background area information; S2: Responding to the axial movement of the endoscope, capturing narrowband light images at a first preset position and a second preset position of the endoscope away from the intestinal wall during the axial movement, calculating brightness gain values ​​of a blue light channel and a green light channel based on the narrowband light images at the first preset position and the second preset position, calculating an asymmetric scattering correction coefficient based on the brightness gain values ​​of the blue light channel and the green light channel, and updating an indication of color information calibration completeness based on the stability of the asymmetric scattering correction coefficient; S3: responding to the advancement operation of the endoscope, tracking the pixel movement speed of the interference object in the field of view during the advancement operation, calculating the relative depth of the interference object according to the pixel movement speed of the interference object, obtaining an interference layer depth distribution map, and updating the indication of the validity of the interference layer depth data according to the validity of the interference layer depth distribution map; S4: In response to the scanning operation of the endoscope, identifying the image area not blocked by the interference object, storing the image information of the image area not blocked by the interference object in the background canvas, and updating the indication of the background area information coverage according to the coverage of the background canvas; S5: When the indicated color information calibration completeness, interference layer depth data validity, and background area information coverage are all achieved, triggering the image synthesis operation; S6: performing color compensation on the image information of the image area not blocked by the interference object stored in the background canvas according to the asymmetric scattering correction coefficient; S7: Combine the interference layer depth distribution map and the background canvas to reconstruct a clear image; S8: Perform lesion detection on the clear image.

2. A fiber endoscope image lesion detection method as claimed in claim 1, characterized in that: Step S2 includes: In response to the axial movement of the endoscope, continuously capturing a sequence of images for color correction during the axial movement; For each frame of the image sequence for color correction, a central area and an edge area are divided; Calculate the brightness changes of the blue light channel and the green light channel in the center area and the brightness changes of the blue light channel and the green light channel in the edge area respectively; Compare the brightness changes of the blue light channel in the central area with those in the edge areas, and compare the brightness changes of the green light channel in the central area with those in the edge areas to identify whether there is a particle crowding effect in the central area; The brightness change of the blue light channel in the edge area is used as the true brightness change of the blue light channel reflecting the inherent optical attenuation characteristics under the body fluid interference environment, and the brightness change of the green light channel in the edge area is used as the true brightness change of the green light channel reflecting the inherent optical attenuation characteristics under the body fluid interference environment; Calculate the brightness gain value of the blue light channel according to the actual brightness change of the blue light channel; Calculate the brightness gain value of the green light channel according to the actual brightness change of the green light channel; Calculate the asymmetric scattering correction coefficient according to the brightness gain values ​​of the blue light channel and the green light channel; The indication of the completeness of the color information calibration is updated based on the stability of the asymmetric scatter correction coefficient.

3. The method for detecting lesions in fiber endoscope images according to claim 1, wherein: Some of the steps in step S3 include: In response to the advancing operation of the endoscope, continuously capturing an image sequence for depth estimation during the advancing operation; From the image sequence used for depth estimation, identify the image region where the distractor is located in the field of view; For the image area where the interference object is located, the motion vector of the pixel point in the image area where the interference object is located is estimated between consecutive frames of the image sequence used for depth estimation; According to the motion vector of the pixel point, the pixel movement speed of the interference object is calculated.

4. The method for detecting lesions in fiber endoscope images according to claim 1, wherein: Some of the steps in step S3 include: In response to the advancement operation of the endoscope, a compound detection action is performed, wherein the compound detection action includes a uniform advancement phase and a momentary pause phase; In the instantaneous pause stage, the particles in the interference are classified into first inertia particles and second inertia particles according to the motion velocity attenuation characteristics of the particles in the interference; In the uniform advancing stage, the relative depth of the first inertial particle is calculated according to the pixel moving speed of the first inertial particle; According to the relative depth of the first inertial particle, the distance to the intestinal mucosa is determined; According to the distance to the intestinal mucosa, the spatial area between the endoscope tip and the intestinal mucosa is determined as the depth range of the fluid layer; Setting the relative depth of the second inertial particle to a preset depth or depth range within the depth range of the fluid layer; Integrating the relative depth information of the first inertial particles and the relative depth information of the second inertial particles to generate an interference layer depth distribution map; The indication of validity of the interference layer depth data is updated according to the validity of the interference layer depth distribution map.

5. The fiber endoscope image lesion detection method according to claim 1, characterized in that: Step S4 includes: In response to a scanning operation of the endoscope, continuously capturing a sequence of images for background area accumulation during the scanning operation; For each frame of the image sequence used for background area accumulation, a local area that meets the definition requirement in the frame is identified to obtain a clear area; Evaluate the optical uniformity of each clear area to quantify its residual interference level and generate corresponding clarity quality parameters; The image information of each clear area and its clarity quality parameters are stored together; Perform image registration on the image information of the clear area, and accumulate the registered image information to the background canvas; During the accumulation of the registered image information, the image information corresponding to the clear area position stored in the background canvas is compared with the image information stored at the corresponding position, and based on the values ​​of their respective clarity quality parameters, the image information with the higher clarity quality parameter value is selected for updating or merging into the background canvas; The coverage of the background canvas is calculated according to the ratio of the number of valid pixels filled in the background canvas to the total number of pixels, and the indication of the coverage of the background area information is updated.

6. A fiber endoscope image lesion detection method as claimed in claim 5, characterized in that: The steps for evaluating each clear area for optical uniformity to quantify its residual interference level include: dividing the clear area into a plurality of sub-areas; Capture the image sequence corresponding to the clear area; For each sub-region in the image sequence corresponding to the clear area, the brightness change, color change and texture change of the pixels in each sub-region are calculated to evaluate the optical uniformity of each sub-region; Generate corresponding residual interference metrics based on the brightness changes, color changes, and texture changes of pixels in each sub-region; Based on the residual interference metrics of all sub-regions, the residual interference degree of the clear area is comprehensively quantified.

7. A fiber endoscope image lesion detection method according to claim 6, characterized in that: For each sub-region in the image sequence corresponding to the clear region, the steps of calculating the brightness change, color change, and texture change of pixels in each sub-region include: Obtaining brightness information of pixels in the sub-region in a time dimension, and calculating brightness fluctuation amplitudes of pixels in the sub-region based on the brightness information in the time dimension; Obtaining the color information of the pixels in the sub-region in the time dimension, and calculating the color fluctuation amplitude of the pixels in the sub-region based on the color information in the time dimension; The texture information of the pixels in the sub-region in the time dimension is obtained, and the texture fluctuation amplitude of the pixels in the sub-region is calculated based on the texture information in the time dimension.

8. A fiber endoscope image lesion detection method as claimed in claim 7, characterized in that: The steps of generating the corresponding residual interference metric according to the brightness change, color change and texture change of the pixels in each sub-region include: A weighted fusion is performed according to the brightness fluctuation amplitude, color fluctuation amplitude and texture fluctuation amplitude of the pixels in the sub-region to obtain a residual interference measure of the sub-region.

9. The fiber endoscope image lesion detection method according to claim 8, characterized in that: The steps of performing weighted fusion based on the brightness fluctuation amplitude, color fluctuation amplitude, and texture fluctuation amplitude of the pixels in the sub-region to obtain the residual interference measurement of the sub-region include: Determining weights corresponding to brightness fluctuation amplitudes, color fluctuation amplitudes, and texture fluctuation amplitudes of pixels within the sub-region based on preset rules or dynamic adjustment rules; According to the brightness fluctuation amplitude, color fluctuation amplitude, texture fluctuation amplitude and corresponding weights of the pixels in the sub-region, weighted fusion is performed to obtain the residual interference measure of the sub-region.

10. A fiber endoscope image lesion detection system, applied to the fiber endoscope image lesion detection method according to claim 1, characterized in that: The system includes: The data quality guidance module is used to provide a data quality guidance interface and indicate the completeness of color information calibration, the validity of interference layer depth data, and the coverage of background area information; a color correction module, configured to respond to axial movement of the endoscope, capture narrowband light images at a first preset position and a second preset position of the endoscope away from the intestinal wall during the axial movement, calculate brightness gain values ​​of a blue light channel and a green light channel based on the narrowband light images at the first preset position and the second preset position, calculate an asymmetric scattering correction coefficient based on the brightness gain values ​​of the blue light channel and the green light channel, and update an indication of color information calibration completeness based on a degree of stability of the asymmetric scattering correction coefficient; a depth estimation module, configured to respond to the advancement operation of the endoscope, track the pixel movement speed of the interference object in the field of view during the advancement operation, calculate the relative depth of the interference object based on the pixel movement speed of the interference object, obtain an interference layer depth distribution map, and update the indication of the validity of the interference layer depth data based on the validity of the interference layer depth distribution map; a background construction module, configured to respond to a scanning operation of the endoscope, identify an image area not blocked by an interfering object, store image information of the image area not blocked by the interfering object into a background canvas, and update an indication of background area information coverage according to coverage of the background canvas; An image synthesis module is configured to trigger an image synthesis operation when the indications of color information calibration completeness, interference layer depth data validity, and background area information coverage are all met, and perform color compensation on the image information of the image area not blocked by the interference object stored in the background canvas according to the asymmetric scattering correction coefficient, and reconstruct a clear image by combining the interference layer depth distribution map and the background canvas; The lesion detection module is used to perform lesion detection on clear images.

Images