Underwater image processing method for waterlogged bamboo slips and related device

Abstract

The embodiment of the application discloses a kind of underwater image processing method of waterlogged bamboo slips and related device, method includes: the image sequence of bamboo slip collected in water is acquired;To each frame image in N frame image, the preprocessing of cross-media geometric distortion correction and three-dimensional flat field normalization is sequentially executed, and N frame image after preprocessing is obtained;Using the pixel information of corresponding spatial position in N frame image after preprocessing, the backscattering component of N frame image after preprocessing is stripped under the framework of underwater imaging physical model, and N frame despeckling image is obtained;N frame despeckling image is spatially aligned, and through the same spatial position image block comparison, abnormal image block is rejected and fused reconstruction, and the unobstructed image of single frame is obtained;The high-frequency reflection component image of unobstructed image is subjected to local adaptive contrast enhancement processing, and the target bamboo slip image after enhancement is output.The target bamboo slip image ink is clear, and distortion degree is low, and the efficiency and fidelity of bamboo slip digitization collection are improved.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to an underwater image processing method and related apparatus for waterlogged bamboo slips. Background Technology

[0002] As an important ancient document carrier, bamboo and wooden slips contain a wealth of historical, legal, economic, and cultural information. Their excavation environments (such as damp and dark ancient wells and riverbed sediment layers) are generally water-saturated. After excavation, the materials of bamboo and wooden slips are mostly loose bamboo and wood with coarse fibers. If they are directly exposed to the air, they are very prone to warping, cracking, and ink fading. Therefore, the field of cultural relic preservation generally adopts the "water immersion preservation" method, which allows bamboo and wooden slips to be displayed and stored for a long time in a relatively stable water environment. At the same time, it is necessary to preserve the text on the surface of the bamboo and wooden slips in high-fidelity digital form to provide basic data for subsequent text recognition and historical research.

[0003] Currently, there are two main technical approaches to the digital acquisition of bamboo and wooden slips: one is the infrared imaging method based on the air environment. This method requires the waterlogged bamboo and wooden slips to be dehydrated and dried before being photographed. However, the dehydration process will cause irreversible physical damage to the cultural relics, including ink fading, smudging, and deformation of the wooden base, resulting in the permanent loss of valuable textual details. The other is the single-unit sealed underwater acquisition method, which is to take pictures directly in the underwater preservation solution. However, underwater imaging faces serious optical degradation effects: the absorption and scattering of water molecules and suspended particles will cause the light energy to decay rapidly. Traditional point light sources or surface light sources are prone to producing strong backscattering glare and local shadows, resulting in color distortion, false contours, and a sharp drop in contrast in the acquired images, making it difficult to use them directly for text recognition.

[0004] With the large-scale excavation of bamboo and wooden slips in various regions, how to achieve high-fidelity, large-scale digital acquisition of the text on the surface of waterlogged bamboo and wooden slips while avoiding damage to cultural relics has become an urgent technical problem to be solved. Summary of the Invention

[0005] The main objective of this invention is to provide an underwater image processing method and related apparatus for waterlogged bamboo and wooden slips, which can solve the problems of text distortion, low contrast, and difficulty in directly using underwater bamboo and wooden slip images for text recognition in the prior art.

[0006] To achieve the above objectives, the first aspect of the present invention provides an image sequence for acquiring waterlogged bamboo slips collected underwater, the image sequence comprising N frames of images with partially overlapping fields of view, the N frames of images having different viewpoints; For each of the N frames, cross-medium geometric distortion correction and three-dimensional flat field normalization are performed sequentially to obtain the preprocessed N frames. Using the pixel information of the corresponding spatial locations in the preprocessed N frames of images, the backscattering component is removed from the preprocessed N frames of images within the framework of the underwater imaging physical model to obtain N frames of descattered images. Spatial alignment is performed on the N frames of descattered images. Abnormal image blocks are removed by comparing image blocks at the same spatial location and then fused and reconstructed to obtain a single frame of occlusion-free image. The high-frequency reflection component image of the unobstructed image is subjected to local adaptive contrast enhancement processing to output the enhanced target bamboo slip image. The high-frequency reflection component image is used to reflect the ink characteristics of the waterlogged bamboo slip. The high-frequency reflection component image is obtained by decoupling from the limited illumination component image. The limited illumination component image is obtained by constraining the low-frequency illumination component image of the unobstructed image using a pre-calibrated lower illumination threshold. The lower illumination threshold is pre-calibrated based on the lower limit of the physical reflectance of waterlogged wood fibers in the entire wavelength band.

[0007] In one feasible implementation, the step of using pixel information corresponding to spatial locations in the preprocessed N frames of images to remove the backscattering component from the preprocessed N frames of images within the framework of an underwater imaging physical model, to obtain N frames of descattered images, includes: The preprocessed N frames of images are divided into M image blocks; The darkest pixel is extracted from each image block of each frame to form the backscatter intensity vector of each frame, resulting in N backscatter intensity vectors. By comparing the elements of the backscatter intensity vectors of different frames at corresponding spatial positions, the M global minimum values ​​of the M local regions corresponding to the M image blocks are determined. A multi-view backscattering intensity vector is constructed based on the M global minimum values; Based on the multi-view backscattering intensity vector, the coefficients of the artificial illumination distribution model are solved to generate a backscattering component image; Using predetermined water depth parameters, the backscattering component image, and a preset Jaffe-McGlamery model, scattering degradation inversion is performed on the N frames of geometrically corrected images. The backscattering component is stripped pixel by pixel from the N frames of geometrically corrected images to obtain N frames of descattered images. The underwater imaging physical model includes the Jaffe-McGlamery model.

[0008] In one feasible implementation, where N is an odd number and N is greater than 1, the spatial alignment of the N frames of descattered images includes: The middle frame of the N frames of descattered images is used as the reference frame; The feature points of each frame of the descattered image are obtained by using a preset feature detection algorithm to detect feature points in each frame. A preset feature point matching algorithm is used to perform feature point matching processing on the remaining frames other than the reference frame, so as to obtain matching feature point pairs between the remaining frames and the reference frame. Based on the feature point pairs, calculate the perspective transformation matrix of each of the remaining frames relative to the reference frame; Based on the perspective transformation matrix, the images of the remaining frames are transformed to the coordinate system of the reference frame to obtain a spatially aligned image.

[0009] In one feasible implementation, the step of comparing and removing abnormal image blocks at the same spatial location and then fusing and reconstructing them to obtain a single frame of occlusion-free image includes: Extract the completely overlapping common viewing regions from the spatially aligned image and determine them as regions of interest; The region of interest is divided into multiple image blocks; The average intensity difference is calculated by pairing image patches at the same spatial location; Abnormal image blocks with average intensity differences exceeding a preset threshold are removed; and the image block with the smallest average intensity difference is selected and its pixel-wise average is calculated for fusion and reconstruction to obtain a single frame of unobstructed image.

[0010] In one feasible implementation, the step of performing local adaptive contrast enhancement processing on the high-frequency reflection component image of the unobstructed image to output the enhanced target bamboo slip image further includes: The unobstructed image is converted to the logarithmic domain to obtain a logarithmic domain image; The logarithmic domain image is smoothed by a multi-scale Gaussian convolution function to obtain a low-frequency illumination component image; Using the lower limit threshold, the target illuminance values ​​in the low-frequency illuminance component image that are below the lower limit threshold are truncated and raised to the lower limit threshold to generate a restricted illuminance component image. Based on the limited illumination component image and the logarithmic domain image, the high-frequency reflection component image is decoupled.

[0011] In one feasible implementation, the preprocessing of performing cross-medium geometric distortion correction and three-dimensional flat-field normalization sequentially on each of the N frames of images to obtain preprocessed N frames of images includes: The three-dimensional flat-field normalization algorithm, the white field image and the black field image of different color channels are pre-calibrated to perform three-dimensional flat-field normalization processing on each frame of the image pixel by pixel and color channel by color channel to obtain N frames of normalized images. A cross-medium ray tracing model is constructed based on Snell's law. The thickness and refractive index of the waterproof cover glass, the refractive index of the preservation liquid, and the vertical object distance from the camera optical center to the surface of the water-saturated bamboo slip are used as known constant inputs. The ray path from the camera pixel coordinates through each medium layer to the surface of the water-saturated bamboo slip is traced in reverse. The nonlinear offset vector of each pixel in the normalized image of each frame is calculated. Based on the offset vector, spatial coordinate remapping is performed on the normalized images of each frame to obtain a corrected image that eliminates cross-medium distortion. The preprocessed N-frame images include the corrected image.

[0012] To achieve the above objectives, a second aspect of the present invention provides an underwater image processing apparatus for water-saturated bamboo slips, the apparatus comprising: Image acquisition module: used to acquire an image sequence of waterlogged bamboo slips collected underwater, the image sequence including N frames of images with partial field of view overlap, the N frames of images having different viewpoints; Image preprocessing module: used to sequentially perform cross-medium geometric distortion correction and three-dimensional flat field normalization preprocessing on each of the N frames of images to obtain the preprocessed N frames of images; Descattering module: Used to extract the backscattering component from the preprocessed N frames of images by utilizing the pixel information of the corresponding spatial positions in the preprocessed N frames of images within the framework of the underwater imaging physical model, so as to obtain N frames of descattered images; Occlusion removal module: used to spatially align the N frames of descattered images, remove abnormal image blocks by comparing image blocks at the same spatial position and fuse and reconstruct them to obtain a single frame of occlusion-free image; Ink Enhancement Module: Used to perform local adaptive contrast enhancement processing on the high-frequency reflection component image of the unobstructed image, and output the enhanced target bamboo slip image; the high-frequency reflection component image is used to reflect the ink characteristics of the waterlogged bamboo slip, and the high-frequency reflection component image is obtained by decoupling from the limited illumination component image. The limited illumination component image is obtained by constraining the low-frequency illumination component image of the unobstructed image using a pre-calibrated lower illumination threshold. The lower illumination threshold is pre-calibrated based on the lower limit of the physical reflectance of waterlogged wood fibers in the entire wavelength band.

[0013] To achieve the above objectives, a third aspect of the present invention provides an underwater multi-view scanning device for bamboo and wooden slips, the underwater multi-view scanning device comprising: A water tank, the bottom of which is used to hold several bamboo slips, and the water tank is used to provide an underwater environment for collecting images of the bamboo slips; A feed actuator is mounted on top of the water tank; An image acquisition device is mounted on the feed execution component and moves with the feed execution component to acquire images of the bamboo slips at the bottom of the water tank during the feeding process, so as to obtain an image sequence of the water-saturated bamboo slips. A flat-field waterproof cover assembly is fitted over the outside of the image acquisition device. During image acquisition, the bottom of the flat-field waterproof cover assembly extends below or is close to the liquid surface of the water tank. A ring-shaped shadowless soft light illumination array surrounds the water level line attached to the inner wall of the water tank, providing a shadowless environment during image acquisition. The main control unit has a communication connection with the ring-shaped shadowless soft light illumination array, the image acquisition device, and the feed execution component to send working instructions to the ring-shaped shadowless soft light illumination array, the image acquisition device, and the feed execution component. The main control unit is also used to perform the steps shown in the first aspect and any feasible implementation.

[0014] To achieve the above objectives, a fourth aspect of the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps shown in the first aspect and any feasible implementation.

[0015] To achieve the above objectives, a fifth aspect of the present invention provides a computer device including a memory and a processor, the memory storing a computer program, which, when executed by the processor, causes the processor to perform the steps shown in the first aspect and any feasible implementation.

[0016] The embodiments of the present invention have the following beneficial effects: This invention provides an underwater image processing method for waterlogged bamboo and wooden slips. The method includes: acquiring an image sequence of bamboo and wooden slips collected underwater, the image sequence comprising N frames of images with partially overlapping fields of view, the N frames having different viewpoints; performing cross-medium geometric distortion correction and three-dimensional flat-field normalization preprocessing sequentially on each of the N frames to obtain preprocessed N frames; using pixel information corresponding to spatial positions in the preprocessed N frames to remove backscattering components within an underwater imaging physical model framework to obtain N frames of descattered images; performing local adaptive contrast enhancement processing on the high-frequency reflection component image of the unobstructed image to output an enhanced target bamboo and wooden slip image; the high-frequency reflection component image is used to reflect the ink characteristics of the waterlogged bamboo and wooden slips, the high-frequency reflection component image is obtained by decoupling from the limited illumination component image, the limited illumination component image is obtained by constraining the low-frequency illumination component image of the unobstructed image using a pre-calibrated lower illumination threshold, the lower illumination threshold being pre-calibrated based on the lower limit of physical reflectivity of waterlogged wood fibers across the entire wavelength band.

[0017] This invention provides an underwater image processing method for water-saturated bamboo slips. Through multi-view image acquisition combined with a layered processing architecture, it first addresses the cross-medium distortion problem in underwater imaging by constructing a ray tracing model based on Snell's law for nonlinear distortion correction. Simultaneously, it eliminates uneven illumination through three-dimensional flat-field normalization, resulting in a distortion-free, uniformly exposed pre-processed image. Secondly, based on an underwater imaging physical model, it constructs a global backscattering intensity vector using the darkest pixel information from the multi-view images, and removes the backscattering component pixel by pixel, effectively eliminating underwater glare and color distortion. Next, through multi-view image alignment and outlier removal and fusion, it eliminates transient occlusion interference from suspended particles, bubbles, etc., resulting in a clear, unobstructed base image. Finally, considering the material characteristics of the water-saturated bamboo slips, it constrains the low-frequency illuminance component using a pre-calibrated illuminance lower limit threshold based on the physical reflectivity lower limit of water-saturated wood fibers across the entire wavelength band. This decouples the high-frequency reflection component reflecting ink characteristics and performs local adaptive contrast enhancement, suppressing wood texture interference while strengthening the clarity of ink edges. This method eliminates the need for dehydration of waterlogged bamboo and wooden slips, thus avoiding damage to the artifacts. At the same time, the output images of the target bamboo and wooden slips have clear ink marks and low distortion, which can be directly used for text recognition and digital archiving. It supports batch and continuous processing, which greatly improves the efficiency and fidelity of digital acquisition of bamboo and wooden slips. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] in: Figure 1(a) is a schematic diagram of the structure of an underwater bamboo and wooden slips multi-view scanning device according to an embodiment of the present invention; Figure 1(b) is a schematic diagram of the structure of an underwater bamboo and wooden slips multi-view scanning device according to an embodiment of the present invention. Figure 2 ; Figure 1(c) is a schematic diagram of the underwater bamboo and wooden slips multi-view scanning device in an embodiment of the present invention. Figure 3 ; Figure 2 This is a flowchart of an underwater image processing method for waterlogged bamboo slips according to an embodiment of the present invention; Figure 3 This is a structural block diagram of an underwater image processing device for water-saturated bamboo slips according to an embodiment of the present invention; Figure 4 This is a structural block diagram of a computer device in an embodiment of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] This application abandons the traditional single-unit closed shooting mode. Targeting the long and narrow shape and water-saturated preservation characteristics of bamboo and wooden slips, it pioneers a modular underwater high-throughput acquisition device consisting of "top-mounted stepping slide rail scanning + inner wall surround shadowless lighting + non-destructive clamping guide rail". Combined with multi-view synthetic aperture and underwater light field inversion algorithm, it realizes large-scale, high-throughput, and high-fidelity underwater multi-view image acquisition and text restoration of water-saturated bamboo and wooden slips.

[0022] Please refer to Figures 1(a), 1(b), and 1(c). Figure 1(a) is a schematic diagram of the structure of an underwater multi-view scanning device for bamboo and wooden slips according to an embodiment of the present invention; Figure 1(b) is a schematic diagram of the structure of an underwater multi-view scanning device for bamboo and wooden slips according to an embodiment of the present invention. Figure 2 Figure 1(c) is a schematic diagram of the structure of an underwater bamboo and wooden slips multi-view scanning device according to an embodiment of the present invention. Figure 3 Figure 1(b) is a side view of Figure 1(a), and Figure 1(c) is a top view of Figure 1(a). The underwater bamboo slip multi-view scanning device includes: A water tank 10 is provided, the bottom of which is used to hold several bamboo slips 60. The water tank provides an underwater environment for acquiring images of the bamboo slips. Exemplarily, the water tank can be a large-capacity open-type support tank, made of opaque material, with multiple bamboo slips arranged side-by-side in an array at the bottom. The interior is equipped with a flexible slot bracket made of an inert polymer material (such as PEEK) to maintain the bamboo slips' original water-saturated state. That is, a flexible slot bracket can be placed at the bottom of the water tank for loading the bamboo slips.

[0023] A feed execution component 20 is mounted on top of the water tank. Exemplarily, the feed execution component includes a linear displacement execution component. The feed execution component 20 can be a non-destructive, interlocking, high-precision slide rail assembly. This feed execution component 20 consists of an MGN12 standard linear guide rail mounted parallel above the water tank. Utilizing an innovative "corner code clamping" three-dimensional interlocking limiting component (i.e., using the vertical surface of a 304 stainless steel corner code to fix the slider, and the horizontal surface to press the edge of the optical carrier plate), the guide rail slider and the standard optical breadboard are physically assembled without drilling or adhesive. Both ends of the guide rail are equipped with anti-explosion safety limiting modules composed of corner codes and tiger clamps. This slide rail platform allows image acquisition devices 30, such as cameras, to move freely above the water tank to achieve rapid "one-by-one / spanning-column" scanning of the bottom bamboo slip array.

[0024] An image acquisition device 30 is mounted on the feed execution component 20 and moves with the feed execution component to acquire images of the bamboo slips at the bottom of the water tank during the feeding process, thereby obtaining an image sequence of the water-saturated bamboo slips. For example, the image acquisition device 30 can be a camera. Regarding the camera, lens selection, and waterproof optical path: For the requirement of high-precision global acquisition of the bamboo slips, this application selects an industrial camera. The field of view of the industrial camera is sufficient to accommodate at least one bamboo slip, allowing a single photograph to completely cover the bamboo slip while providing a safety margin. This achieves one-click global acquisition in seconds while fully ensuring high-definition recognition of text outlines and ink marks.

[0025] A field-planing waterproof cover assembly 40 is fitted over the image acquisition device 30. During image acquisition, the bottom of the field-planing waterproof cover assembly extends below or is in close contact with the liquid surface of the water tank. For example, the field-planing waterproof cover assembly 40 is an optical field-planing waterproof cover, which is a waterproof cover with a precision-polished high borosilicate glass bottom surface fitted over the camera lens. During shooting, the bottom of the waterproof cover extends below the water surface, completely eliminating dynamic refractive distortion caused by minor fluctuations in the liquid surface on large field-of-view imaging.

[0026] A ring-shaped shadowless soft light illumination array 50 is installed below the water level line on the inner wall of the water tank, providing a shadowless environment during image acquisition. For example, the ring-shaped shadowless soft light illumination array 50 is an underwater ring-shaped shadowless soft light illumination array, consisting of 24V high-density COB flexible light strips laid around the perimeter of the inner wall of the water tank, directly against the wall. The built-in light source directly releases the area array light source underwater, avoiding refraction distortion at the interface between the air and the water surface.

[0027] A main control unit (not shown) is connected to the annular shadowless soft light illumination array, the image acquisition device, and the feed execution component to send operating instructions to them. The main control unit is also used to execute the steps of the underwater image processing method for water-saturated bamboo slips proposed in this application, i.e., to execute... Figure 2 The steps of the method are as follows. For example, when image acquisition begins, the main control unit sends a driving light emission command to the illumination array, a feeding movement command to the feeding execution component, and an acquisition start command to the image acquisition device.

[0028] The underwater bamboo slips multi-view scanning device also includes a slider 21, which is mounted on the feed execution component 20 and moves on it. The feed execution component 20 moves above the water tank, and the slider 21 moves on the feed execution component, allowing the image acquisition device 30 to move both horizontally and vertically to scan the bamboo slips at the bottom. Specifically, the slider 21 can be a vertically fixed slider made of 304 stainless steel corner brackets.

[0029] This application constructs a specific "light source-medium-target-camera" structured underwater image projection model based on the physical characteristics of the water-saturated bamboo slips, which have a highly reflective water film on their outer surface, a slender body, and microscopic interlocking cracks.

[0030] This application employs an active underwater non-direct panoramic light field evolution mechanism. During the digitization process, waterlogged bamboo slips must be completely immersed in the preservation solution. Because the wooden substrate is covered with a non-uniform, continuous water film, conventional external direct light sources or single-point underwater direct light produce strong specular reflections (glare), causing local ink information to be physically blocked due to overexposure. Simultaneously, the unidirectional nature of direct light leaves deep shadows on the warping, deformation, and cracks in the wood of the bamboo slips. This application achieves closed-loop lighting by attaching a ring-shaped soft light array below the water level line on the inner wall of an open-type water tank. The linear or area array excitation light emitted by the light source, when passing through the complex preservation solution containing fine suspended particles, fully excites and utilizes the Mie scattering effect of the medium itself. The light entering the liquid evolves through multiple disordered scattering and spatial overlap by the particles, forming a 360° isotropic, non-directional, three-dimensional uniform diffuse light field at the bottom of the tank. The light field completely envelops the surface of the bamboo and wooden artifacts submerged in the preservation liquid, suppressing the specular reflection component of the surface water film from the physical light path source, and completely eliminating the background shadow at local depressions and cracks in the material, creating an ultimate underwater shadowless environment.

[0031] Simultaneously, by utilizing the refractive compression of a large field-of-view flat-field image projection constraint, the precision glass plane at the bottom of the flat-field waterproof cover assembly surrounding the lens of the top-mounted industrial camera slightly protrudes below the water surface or closely adheres to the liquid surface when the camera moves along the linear displacement actuator. This structure uses a physically rigid plane to forcibly flatten the gently rippling gas-liquid interface, eliminating random air-water refractive distortion caused by dynamic fluctuations in the liquid surface.

[0032] For example, one system hardware configuration of this application is as follows: a highly modular, high-throughput hardware system designed for "mass production of long, narrow (approximately 25cm) waterlogged bamboo slips," mainly composed of the following modules: Large-capacity open-type water tank: Made of opaque material, with external dimensions of 60cm long, 40cm wide, and 45cm high. The bottom of the tank (approximately 400mm deep) can accommodate multiple bamboo slips, each 20-25cm long, arranged side-by-side. Internally, it features flexible slotted brackets made of inert polymer material (such as PEEK) to maintain its original water saturation.

[0033] Non-destructive interlocking high-precision slide rail assembly: An MGN12 standard linear guide rail is mounted parallel above the water tank. Utilizing an innovative "corner code clamping" three-dimensional interlocking limiting assembly (i.e., using 304 stainless steel corner codes to fix the slider vertically and press the optical carrier plate edge horizontally), the guide rail slider and standard 100×100mm optical breadboard are physically assembled without drilling or adhesive. Both ends of the guide rail are equipped with anti-push-out safety limiting modules consisting of corner codes and tiger clamps. This slide rail platform allows the camera to move freely above the water tank, enabling rapid "single / spanning" scanning of the bottom bamboo slip array.

[0034] Underwater Circular Shadowless Soft Light Array: A 24V high-density COB flexible light strip is laid around the perimeter of the tank, below the water level line and closely adhering to the wall. The built-in light source directly emits an array of light sources underwater, avoiding refraction distortion at the interface between the air and water surface.

[0035] Camera, Lens Selection, and Waterproof Optical Path: To address the requirement for high-precision global acquisition of bamboo and wooden slips, this solution utilizes the Daheng Imaging MER2-1220-32U3M / C industrial area array camera (12.2 megapixels, 1 / 1.7-inch sensor surface, 4024×3036 resolution, sensor width approximately 7.44mm). To ensure a single image completely covers the bamboo and wooden slips while allowing for safety margins, this solution prioritizes an industrial fixed-focus lens with a focal length of f=8mm. With this configuration, the physical accuracy per pixel reaches approximately 280mm / 4024≈69μm / pixel, achieving one-click global acquisition in seconds while fully guaranteeing high-definition recognition of text outlines and ink marks.

[0036] Optical Plane Waterproof Cover: A waterproof cover with a precision-polished (Ra<0.05μm) high borosilicate glass bottom surface is fitted over the lens. During shooting, the bottom of the waterproof cover is submerged below the water surface, completely eliminating dynamic refraction distortion caused by tiny fluctuations in the liquid surface on the large field of view imaging.

[0037] Furthermore, the overall operation and usage of the device are as follows: Based on the above modular device, the high-throughput digital acquisition process for batch bamboo and wooden slips is as follows: Open-style exhibition: A large number of waterlogged bamboo slips are placed flat in the brackets at the bottom of the water tank, and deionized water or preservation solution is poured in to a height of 40cm. A standard high-reflectivity 24-color card and a waterproof scale are placed on the side.

[0038] Black and white field calibration: Turn on underwater ring COB illumination to acquire black and white field reference images to eliminate sensor dark current and light field background deviation.

[0039] High-throughput array acquisition: Utilizing the 280mm wide field of view provided by the 8mm lens, the camera is triggered to perform a full-frame exposure, completing the digital preservation of a single bamboo slip in a single shot. Subsequently, the camera is manually or electrically driven to slide along a linear guide rail to be directly above the next bamboo slip, and the exposure is repeated.

[0040] Seamless transfer: No vacuuming or separate sealing of the bamboo slips is required throughout the process. The water tank serves as the production line workbench, significantly reducing the exposure time of the bamboo slips after they are removed from the original protective liquid.

[0041] This application provides an underwater multi-view scanning device for bamboo and wooden slips, which has the following beneficial effects: First, it significantly improves the throughput and operational efficiency of acquiring large batches of bamboo and wooden slips: This invention constructs a large field of view covering an area of ​​approximately 280mm through precise optical path matching. This optical path design overcomes the technical bottleneck of traditional micro-acquisition, which relies on multiple step-by-step translations by mechanical mechanisms and subsequent stitching of multiple frames of images, and realizes global one-time exposure imaging of single or even multiple arrays of bamboo and wooden slips. It greatly shortens the exposure time of bamboo and wooden slips after they are removed from their original protective liquid, fundamentally improving the engineering throughput of digital acquisition of massive amounts of unearthed bamboo and wooden slips; Second, it achieves high-rigidity non-destructive assembly and low-cost modularization of general mechanical components: The three-dimensional staggered limiting non-destructive transfer component created by this invention uses right-angle limiting parts (corner codes) to perform multi-directional physical envelope pressing on the edge of the optical carrier plate, effectively overcoming the technical problem of incompatibility between the hole spacing of the standard linear guide slider and the optical carrier plate. The entire process does not require destructive processing such as drilling of standard parts, and also avoids the risk of fatigue failure caused by the use of adhesives in humid environments. While ensuring the high rigidity and smooth sliding of the optical acquisition system, the invention significantly reduces the costs of equipment customization, setup, and maintenance. Thirdly, it effectively suppresses optical degradation effects in slightly turbid water environments, achieving high-fidelity imaging: This invention breaks away from the conventional underwater direct illumination pattern by setting up a ring-shaped illumination array on the inner wall of an open water tank. Utilizing the particle diffuse scattering effect of the slightly turbid water medium itself, a uniform, shadowless light field covering the bottom is constructed. Combined with a top-mounted flat-field optical waterproof cover and the Jaffe-McGlamery physical inversion algorithm at the back end, it eliminates underwater backscattering glare and localized shadows on the surface of cultural relics from both physical optical path acquisition and digital computation dimensions, providing high signal-to-noise ratio and high-fidelity underlying data support for the extraction of faint ink marks on the surface of bamboo slips.

[0042] Furthermore, regarding the image processing and calculation model for the characteristics of waterlogged bamboo and wooden slips, it should be noted that after acquiring a global high-resolution image, this application uses a dedicated intelligent processing module to perform restoration calculations based on the material properties and underwater degradation mechanisms of the bamboo and wooden slip artifacts. This results in high-fidelity images of bamboo and wooden slips with clear ink marks, low distortion, and direct applicability for text recognition and digital archiving, supporting batch continuous processing. In other words, this application proposes a dedicated intelligent processing module that deeply integrates the three-dimensional spatial characteristics of underwater structured optical paths with the specific physical properties of waterlogged wood, achieving non-standardized customized calculation steps. Please refer to the following for details.

[0043] Please see Figure 2 , Figure 2This is a flowchart of an underwater image processing method for waterlogged bamboo slips according to an embodiment of the present invention, as shown below. Figure 2 The method shown includes the following steps: 201. Obtain an image sequence of water-saturated bamboo slips collected underwater, the image sequence comprising N frames of images with partial field-of-view overlap, the N frames of images having different viewpoints; It is understandable that the image sequence of waterlogged bamboo slips acquired underwater is based on the aforementioned underwater multi-view scanning device for bamboo slips. More specifically, it is acquired during the feeding process of the image acquisition device scanning the bamboo slips. Because the image acquisition device moves during the scanning process, it can acquire continuous images of bamboo slips with partial field-of-view overlap, and each frame has a different viewpoint. These multiple frames with different viewpoints are then used for subsequent image processing to achieve a high-fidelity final image of the bamboo slips. The image is the raw two-dimensional pixel matrix (Raw Image) acquired underwater by the camera without any processing, denoted as... ,in, () represents the original pixel coordinates of the image.

[0044] 202. For each frame of the N frames, perform cross-medium geometric distortion correction and three-dimensional flat field normalization preprocessing in sequence to obtain the preprocessed N frames. After obtaining the above N frames of images, we can begin processing them. First, we process each frame of the N frames... The preprocessing steps, namely cross-medium geometric distortion correction and three-dimensional flat-field normalization, are performed sequentially to obtain N preprocessed frames of images. It should be noted that step 202 is based on geometric distortion and background correction of the elongated shape feature. Specifically, the original image is read in. The reflection intensity is normalized and corrected by combining the calibrated black / white field. To address the issue of barrel distortion at the edges of the 8mm wide-angle lens, which causes long bamboo slips to bend and deform at the image edges, nonlinear geometric distortion inverse correction is performed based on the pre-calibrated camera intrinsic parameter matrix to forcibly straighten the edges of the bamboo slips, providing a strictly orthogonal geometric basis for the subsequent digital compilation and splicing of the bamboo slips.

[0045] In one feasible implementation, the preprocessing of performing cross-medium geometric distortion correction and three-dimensional flat-field normalization sequentially on each of the N frames of images to obtain the preprocessed N frames of images includes the following steps A01 to A03: A01. Using a preset three-dimensional flat-field normalization algorithm, pre-calibrated white field images and black field images of different color channels, perform pixel-by-pixel and color channel-by-color three-dimensional flat-field normalization processing on each frame of the image to obtain N frames of normalized images. This step primarily involves three-dimensional flat-field normalization to address the exponential decay of the underwater artificial light field. Unlike conventional two-dimensional brightness stretching, conventional flat-field correction cannot handle the depth-induced attenuation of underwater light sources. This application performs three-dimensional volumetric light field relative reflectivity normalization in step A01. Since the background light from the artificial ring light source at infinity underwater is closely related to the camera-light source distance, and the light intensity decays non-linearly with depth, during the calibration phase, the system submerges a high-reflectivity standard ceramic white plate at a depth of 400mm, identical to that of the bamboo slips, to acquire white field images. ; Read in the original image With black screen Then, using the formula: (1) By eliminating the camera's dark current noise, the edge attenuation of the 8mm wide field-of-view lens, and the nonlinear three-dimensional energy attenuation gradient of the underwater parallel ring light source at a specific water depth, a true reflectivity matrix that is accurately stripped of interference from uneven spatial distribution of the light field is obtained. In the formula, x and y represent the positions in the image matrix (e.g., column x and row y). This indicates that flat-field correction is calculated "pixel-by-pixel" because the attenuation of the underwater light field is completely different at different locations in the image (center and edges); λ: the wavelength of the light wave. In a conventional color area array industrial camera, λ actually represents the camera's three different color channels: R (red), G (green), and B (blue). This is because the attenuation patterns of different wavelengths of light underwater are completely different. Red light attenuates extremely quickly in water, while blue and green light have strong penetrating power. Therefore, three-dimensional volumetric light field normalization cannot simply convert the image to grayscale for uniform processing; it must perform independent division operations on the white and black field data of the corresponding channels for the R, G, and B channels (i.e., different λ bands). This is the core reason why this algorithm can simultaneously correct underwater color cast.

[0046] For example, the system reads in the absolute original image taken directly. First, apply formula (1) to perform three-dimensional flat-field normalization to obtain an intermediate image with uniform illumination but still distorted shape. The complete meaning of formula (1) is: the relative true reflectance of a pixel located at coordinates (x, y) in a color channel with wavelength λ after removing background light field interference. This is equal to the original pixel value at this wavelength minus the black level value, then divided by the corresponding net white level value. Then, this... The image is then fed into the ray tracing model in the next step to remap the pixel coordinates, ultimately resulting in an orthophoto that is both straightened and evenly exposed.

[0047] A02. Based on Snell's law, a cross-medium ray tracing model is constructed. The thickness and refractive index of the waterproof cover glass, the refractive index of the preservation liquid, and the vertical object distance from the camera optical center to the surface of the water-saturated bamboo slip are used as known constant inputs. The ray path from the camera pixel coordinates through each medium layer to the surface of the water-saturated bamboo slip is traced in reverse. The nonlinear offset vector of each pixel in the normalized image of each frame is calculated. A03. Perform spatial coordinate remapping on the normalized images of each frame according to the offset vector to obtain a corrected image that eliminates cross-media distortion. The preprocessed N-frame images include the corrected image.

[0048] Furthermore, geometric distortion correction based on a cross-medium refraction physical model is performed through steps A02 and A03. It should be noted that conventional image distortion correction only addresses the radial / tangential distortion of the lens itself. However, in this application, the light beam must continuously pass through three media with different refractive indices: air (inside the waterproof cover), precision borosilicate glass, and a slightly turbid artifact preservation solution (n≈1.33). The image processing algorithm in this application not only calls the camera intrinsic parameter matrix but also uses the glass thickness, the refractive index of the preservation solution, and the 400mm water depth object distance as joint physical constraints to construct a cross-medium ray tracing distortion model. By calculating the nonlinear refraction offset vector of the beam at the cross-medium interface, a remapping of spatial coordinates is performed, forcibly restoring the true orthogonal contour of the long, narrow bamboo slip, completely eliminating the false geometric stretching caused by water refraction.

[0049] Among them, the cross-medium ray tracing distortion model is the core mathematical framework of the entire correction algorithm. Based on geometric optics (Snell's Law), it reverse-engineers the complete three-dimensional physical path of light rays starting from the camera chip, passing through the lens, air, a waterproof glass cover, and finally entering the water to reach the surface of the bamboo slip.

[0050] Joint physical constraints: These are the input boundary conditions for constructing the above tracking model. Since the refractive offset is not random, but strictly determined by the refractive index, thickness, and distance of the medium, these physical quantities are input into the model as known constants.

[0051] The nonlinear refraction offset vector (also known as the nonlinear offset vector) is a result / intermediate variable calculated by the model. In a wide-angle lens, the angle of incidence of light rays at the edge of the field of view is large, resulting in a deeper degree of refraction and bending; while the angle of incidence of light rays at the center of the field of view is small, resulting in a shallower bending. This inconsistent characteristic of light ray offset at different positions is what makes it "nonlinear." The spatial difference between the theoretical point of refraction and the actual point of refraction calculated by the model is the offset vector.

[0052] Spatial coordinate remapping: This is the final step. After obtaining the offset vector for each pixel across the entire field of view, the algorithm remaps the original image... The pixels in the image are forcibly "moved" to the corrected image according to vector guidance. The image pixels are rearranged at the correct physical orthogonal positions.

[0053] Among these parameters, the glass thickness, the refractive index of the preservation fluid, and the 400mm water depth-object distance are all predetermined, known constants. One of the core aspects of this invention is the use of a "rigid hardware structure" to fix dynamic variables as static parameters. Specifically, this includes: Refractive index of medium: air refractive index ≈1.0; High borosilicate glass refractive index of the flat-field waterproof cover ≈1.47; Refractive index of slightly turbid cultural relic preservation solution ≈1.33 (can be pre-determined by a refractometer).

[0054] Thickness and object distance parameters: Physical thickness of the waterproof cover glass (e.g., 5mm); the perpendicular object distance from the optical center of the camera lens to the surface of the bamboo slip. It is rigidly fixed by mechanical slide rails and water tank height, approximately 400mm.

[0055] Camera intrinsic parameters: camera intrinsic parameter matrix K and lens radial / tangential distortion coefficients ( , , , The data is obtained and cured in advance through air calibration before the equipment leaves the factory.

[0056] In one feasible implementation, steps A02 and A03 can be further divided into four sub-steps, as shown in the example below: Step 1: 3D ray transformation of pixel coordinates and lens distortion pre-compensation processing: For the corrected target image For any desired orthogonal coordinate point (x', y'), the inverse intrinsic parameter matrix of the camera is used. Convert it to a 3D direction vector in the normalized camera coordinate system Simultaneously, combined with the pre-calibrated lens distortion coefficient ( , , , The Brown-Conrady inverse correction is performed on the vector to eliminate the lens’s own mechanical and optical distortion.

[0057] Step 2: Perform Snell's Ray Tracing across media: With the camera's optical center as the origin, Let be the incident direction of the light ray within the air cavity of the waterproof cover, tracing downwards to intersect at the "air-glass" interface. Calculate the refractive direction vector of the light ray within the glass using Snell's law in vector form. : (2) (in This is the vertically downward interface normal vector. (where the angle of incidence is 0).

[0058] Light along Passing through a known thickness After the glass is filled, it is placed at the "glass-preservation solution" interface; substitute the above formula again (to... , Replace with , (This is used to obtain the final direction vector of the light ray underwater.) .

[0059] Step 3: Calculation of nonlinear refraction offset vector: Light along Continue downwards to the known water depth Calculate the actual intersection points of the light rays on the physical plane of the bamboo slip. The actual intersection point Intersection with the theoretical point where no refraction occurs (assuming the entire path is air). The difference between them is the nonlinear refraction offset vector at that specific field of view. - .

[0060] Step 4: Spatial coordinate remapping based on bilinear interpolation: Based on the obtained Backprojection mapping back to the original acquired image The calculated (x, y) values ​​correspond to the floating-point pixel coordinates. Since the calculated (x, y) values ​​are usually non-integer, the algorithm extracts the grayscale values ​​of the surrounding four integer pixels using bilinear interpolation or bicubic interpolation, performs a weighted calculation, and assigns the result to the target image. The above operation is performed on all pixels of the entire image, and the final output is a completely straightened orthophoto of the bamboo slips with cross-media distortion eliminated.

[0061] 203. Using the pixel information of the corresponding spatial positions in the preprocessed N frames of images, the backscattering component is removed from the preprocessed N frames of images under the framework of the underwater imaging physical model to obtain N frames of descattered images. After preprocessing, step 203 performs scattering degradation inversion based on multi-view image patch decomposition. Specifically, general single-image dehazing algorithms heavily rely on deep-water background areas in the image to calculate global light intensity. However, in this high-throughput scanning application, the bamboo slip array almost fills the field of view, and the single-image algorithm will fail due to misidentifying the dark areas of the bamboo slips as background. Therefore, this application utilizes the pixel information of corresponding spatial locations in the preprocessed N-frame images to remove the backscattering component from the preprocessed N-frame images within the framework of the underwater imaging physical model, obtaining N-frame descattered images, thus overcoming the distortion problem of single-image blind estimation.

[0062] It should be noted that the processing object in step 203 is the normalized image obtained after the geometric remapping and 3D flat-field normalization processing in step 202. Only after completely eliminating the mechanical distortion of the lens, the dark current of the sensor, and the spatially uneven attenuation of the ring light source itself, can the dark pixels retained in the image truly represent the backscattering produced by suspended matter in the water.

[0063] For example, single-image algorithms can fail due to misidentifying dark areas of the bamboo slips as background. This application utilizes an N-frame multiview overlapping image sequence acquired by linear guide rail stepping. The algorithm divides the multiview images into m×m image blocks and extracts the darkest pixels from the corresponding image blocks of each viewpoint to form a backscattering intensity vector. Since the hardware guide rail ensures consistent illumination between adjacent viewpoints, the algorithm extracts the globally minimum intensity pixel from the vector as the true backscattering reference for that local area, which is used to solve for the coefficients of the two-dimensional polynomial model f(X,a) of artificial illumination distribution. Subsequently, the hardware absolute parameters (liquid layer thickness d=400mm) are substituted as known constants into the improved Jaffe-McGlamery physical model to perform inversion, achieving accurate physical stripping of the "fog" in the slightly turbid preservation liquid and overcoming the distortion problem of single-image blind estimation.

[0064] In one feasible implementation, the step of using pixel information corresponding to spatial locations in the preprocessed N frames of images to strip the backscattering component from the preprocessed N frames of images within the framework of an underwater imaging physical model, to obtain N frames of descattered images, includes steps B01-B06: B01. Divide the preprocessed N frames of images into M image blocks; B02. Extract the darkest pixel from each image block of each frame to form the backscatter intensity vector of each frame, and obtain N backscatter intensity vectors. B03. By comparing the elements of the corresponding spatial positions in the backscatter intensity vectors of different frames laterally, the M global minimum values ​​of the M local regions corresponding to the M image blocks are determined. B04. Construct a multi-view backscattering intensity vector based on the M global minimum values; B05. Based on the multi-view backscattering intensity vector, solve for the coefficients of the artificial illumination distribution model to generate a backscattering component image. ; B06. Using predetermined water depth parameters, the backscattering component image, and a preset Jaffe-McGlamery model, perform scattering degradation inversion on the N frames of geometrically corrected images, and remove the backscattering component pixel by pixel from the N frames of geometrically corrected images to obtain N frames of descattered images; the underwater imaging physical model includes the Jaffe-McGlamery model.

[0065] The linear guide rail, through equidistant stepping, acquires a series of images with fixed physical displacement and high field-of-view overlap. Specifically, the odd-numbered frame sequence generated by the guide rail stepping can be selected, ensuring that a certain physical pixel on the reference frame can be observed from both a left-hand and a right-hand perspective. This eliminates perspective occlusion bias from a single-sided viewpoint and provides rigorous physical spatial data for finding the "globally darkest pixel."

[0066] A single image (i.e., a certain viewpoint) is divided into M m×m image blocks, and then the darkest pixel is extracted from each image block. These M darkest pixels are combined to form a backscattering intensity vector belonging to that viewpoint (this vector contains M elements).

[0067] The number of viewpoints and vectors are equal: since there are N viewpoints (i.e. N frames of images), a total of N backscattering intensity vectors will be extracted.

[0068] Global minimum intensity pixel (i.e., the pixel corresponding to the global minimum): The algorithm does not select a minimum value for each viewpoint. Instead, for the central reference frame, the algorithm horizontally compares the elements at corresponding spatial positions in these N vectors and extracts the absolute minimum value among these N values. After traversing and comparing, M global minimum values ​​are finally extracted and merged into a final reference frame backscatter intensity vector (i.e., multi-view backscatter intensity vector). This final vector integrates the advantages of multiple viewpoints, eliminates the interference of black ink marks on bamboo slips from a single viewpoint, and represents the true backscatter intensity of the local water area.

[0069] Furthermore, a final vector containing M true backscattering values ​​is obtained. Substituting these M intensity values ​​and their corresponding center coordinates (x, y) into the polynomial forms a system of linear equations with six unknowns. Since the number of image patches M is much greater than 6, this is an overdetermined system of equations. The algorithm solves this system of equations using singular value decomposition (SVD) or least squares fitting to find the coefficient matrix 'a' that best matches the current underwater light field. After the coefficients are determined, substituting all the (x, y) coordinates of the entire image into the formula generates a full-resolution, smooth backscattering component image. .

[0070] The core idea of ​​the Jaffe-McGlamery underwater imaging model is: the final image captured by the camera (i.e., the previous image) This is equivalent to the "direct transmitted light (Scene Radiance)" from the bamboo slip target penetrating the water layer and reaching the camera, plus the "backscattered light" from suspended particles in the water. The purpose of the inversion is to convert the former (the actual radiance of the bamboo slip) into the latter. It is extracted from the equation in reverse.

[0071] The specific execution process of step B06 is as follows: 1) Physical subtraction (fog removal): From the normalized image In the middle, the final backscattering component calculated and fused with ambient light from the previous step is subtracted pixel by pixel. This step physically removed the "luminescent fog of water" that was blocking the bamboo slips.

[0072] 2) Energy Compensation (Substituting Hardware Parameters): The energy of the actual light rays from the bamboo slips decreases exponentially with distance as they pass through water. According to the underwater transmittance formula... The image after subtraction must be divided by the transmittance for compensation.

[0073] 3) Calculation Output: Traditional algorithms here rely on guesswork or approximation using dark channels, resulting in significant errors. In contrast, this application directly uses the absolute water depth parameter d=400mm (locked by the mechanical structure) and the pre-calibrated preservation fluid attenuation coefficient. Substitute it as an absolute truth value. Mathematical expression: (3) The true surface reflectance J(x,y) of the bamboo slips, unaffected by optical interference from water, is calculated by reverse calculation using the above formula, thus completing the inversion. In the formula, That is .

[0074] 204. Spatial alignment is performed on the N frames of descattered images. Abnormal image blocks are removed by comparing image blocks at the same spatial location and then fused and reconstructed to obtain a single frame of unobstructed image. Furthermore, the N frames of descattered images are spatially aligned. Abnormal image blocks are eliminated by comparing image blocks at the same spatial location, and then fused and reconstructed to obtain a single frame of unobstructed image. Microturbid preservation solutions often contain floating sawdust and impurities, causing macroscopic physical occlusion in the images. Step 204 involves extracting suspended matter components and eliminating occlusion based on multi-view spatial coherence. Because the hardware acquires multi-view sequences with high overlap, the visual state of the same bamboo slip's physical coordinates differs under different views. The algorithm extracts the region of interest (ROI) for each aligned frame and introduces the minimum average intensity difference (LPMD) method: pairing corresponding image blocks extracted from different views to calculate the average intensity difference; eliminating abnormal blocks with excessive intensity differences due to suspended matter occlusion, and selecting only high spatial coherence image block pairs with minimal intensity differences and averaging them to reconstruct clear, unobstructed image blocks for the current region, ensuring the integrity of feature extraction from the source.

[0075] In one feasible implementation, N is an odd number, and N is greater than 1. That is, the above N frames of images can be an odd-numbered multiview overlapping image sequence obtained by linear guide stepping. Then, the spatial alignment of the N frames of descattered images includes steps C01-C05: C01. Use the middle frame of the N frames of descattered images as the reference frame; C02. Use a preset feature detection algorithm to detect feature points in each frame of the descattered image to obtain the feature points of each frame; C03. Use a preset feature point matching algorithm to perform feature point matching processing on the remaining frames other than the reference frame with the reference frame to obtain matching feature point pairs between the remaining frames and the reference frame. C04. Calculate the perspective transformation matrix of each of the remaining frames relative to the reference frame based on the feature point pairs; C05. Based on the perspective transformation matrix, transform the images of the remaining frames to the coordinate system of the reference frame to obtain a spatially aligned image.

[0076] Among them, "odd frames" refers to a total number of image frames that are odd (for example, N=3 or 5 frames of images captured during continuous scanning of the slide rail to form a sequence set).

[0077] The purpose of spatial alignment is to obtain aligned frames, that is, to align the same position of the same bamboo slip in different frames. Because the camera undergoes physical displacement on the slider, although the same bamboo slip is being photographed, the pixel coordinates of the bamboo slip in different frames have shifted relatively. Therefore, strict coordinate spatial alignment must be performed through feature matching and perspective transformation (Homography). The specific processing procedure is as follows: 1) Establishing a reference frame: After reading the odd-numbered frame multi-view sequence, the algorithm directly selects the middle view image of the sequence as the absolute reference frame.

[0078] 2) Feature point detection: Using feature detection algorithms such as Accelerated Robust Features (SURF), local feature key points are detected and described in the reference frame and other frames respectively.

[0079] 3) Feature point matching: Using matching algorithms such as Fast Approximate Nearest Neighbor (FLANN), find feature point pairs that match each other between the reference frame and the other frames.

[0080] 4) Calculate the transformation matrix and align: Based on the matched feature points, calculate the perspective transformation matrix of each of the remaining frames relative to the reference frame. Then, use this matrix to perform spatial warping resampling on the other images to make their coordinate systems completely coincide and aligned with the reference frame.

[0081] 5) Extracting overlapping regions (ROI): After alignment, extract the completely overlapping common viewing regions from multiple frames of images and extract them as a unified region of interest (ROI) for use in the next step of occlusion removal.

[0082] In one feasible implementation, the step of comparing and eliminating abnormal image blocks by image blocks at the same spatial location and then fusing and reconstructing them to obtain a single frame of occlusion-free image includes steps D01 to D04: D01. Extract the completely overlapping common viewing regions from the spatially aligned image and determine them as regions of interest; D02. Divide the region of interest into multiple image blocks; D03. Calculate the average intensity difference by pairing image blocks with the same spatial location; D04. Remove abnormal image blocks whose average intensity difference exceeds a preset threshold; and select the image block with the smallest average intensity difference to perform pixel-by-pixel averaging and fusion reconstruction to obtain a single frame of occlusion-free image. .

[0083] Furthermore, to obtain an occluded image, it is necessary to remove occluded aberration image blocks. Specifically, occluded aberration blocks are "discarded," and new blocks are generated using occluded image blocks from other healthy viewpoints. It should be noted that after "spatial alignment," all frames have been transformed to the same coordinate system (specifically, the coordinate system of the selected reference frame). At this point, "same spatial position" refers to the same coordinate position of multiple frames in the same coordinate system.

[0084] The core of this step is the Minimum Patch Average Intensity Difference (LPMD) method, with the following logic: After aligning the same bamboo slip in each image, for the same physical coordinate region, we have N corresponding image patches from different viewpoints. The algorithm performs all pairwise pairings of these N image patches and calculates the average intensity difference for each pair. If there are anomalous patches: if an image patch from a certain viewpoint is occluded by suspended objects in the water, it will inevitably produce a very large intensity difference when paired with another unoccluded image patch. In this case, all pairings that produce large intensity differences are directly discarded / ignored, and only the pair of image patches with the smallest intensity difference is selected. If the intensity difference is extremely small, it means that both image patches from these two viewpoints "see" the real surface of the bamboo slip, are highly similar, and are not occluded. The final new image patch is formed by merging the pixel-by-pixel average intensity difference of these two selected clean image patches. The pixel data of the originally occluded anomalous patch is discarded and does not participate in any calculation.

[0085] Mean Intensity Difference (MID) refers to the average difference in pixel grayscale values ​​between two regions or image blocks, and is used to measure contrast, change detection, or registration error.

[0086] Furthermore, at the level of a single image patch, the reconstructed new patch can be seen as generated by averaging the extracted pair of optimal (unoccluded) image patches. However, at the level of reconstruction of the entire image, there is also a "stitching and recombination" step of spatial array.

[0087] The complete reconstruction process is as follows: Block generation: After selecting the optimal pair for coordinate position p and calculating the mean, a single, absolutely unobstructed new image block is generated.

[0088] Spatial relocation: This generated new image patch will be placed into the final output occluded ROI image (i.e., The corresponding physical coordinate position p.

[0089] Full image traversal stitching: The algorithm traverses all M position intervals divided in the entire image, and for each position, it independently repeats the above-mentioned "pairing -> selecting the pair with the smallest difference -> calculating the mean to generate a new block -> putting it back in its original position".

[0090] Final Formulation: Once all the clean image blocks at these M locations are generated and stitched together into the matrix, a complete and clear image of the bamboo slip, thoroughly free of underwater floating impurities, is reconstructed, resulting in the target bamboo slip image. .

[0091] 205. Perform local adaptive contrast enhancement processing on the high-frequency reflection component image of the unobstructed image to output the enhanced target bamboo slip image; the high-frequency reflection component image is used to reflect the ink characteristics of the waterlogged bamboo slip, the high-frequency reflection component image is obtained by decoupling from the limited illumination component image, the limited illumination component image is obtained by constraining the low-frequency illumination component image of the unobstructed image using a pre-calibrated lower illumination threshold, the lower illumination threshold is pre-calibrated based on the lower limit of physical reflectivity of waterlogged wood fibers in the whole band.

[0092] After processing in steps 201 to 204, an unobstructed image is obtained. Step 205 then performs local adaptive contrast enhancement on the high-frequency reflectance component image R(x,y) of the unobstructed image, outputting the enhanced target bamboo slip image. It should be noted that step 205 introduces a Retinex ink edge enhancement algorithm with a saturated water absorption limit threshold. In contrast, the multi-scale Retinex (MSR) algorithm is mainly used for enhancing bamboo slips dried in air. The wood fibers of saturated bamboo slips absorb moisture, resulting in a significant decrease in reflectivity and deep coupling with the spectrum of faded carbon-based ink (extremely low radiative contrast). Directly applying conventional MSR would incorrectly over-amplify the darkened areas of the water-absorbing wood fiber texture as edge information, producing severe artifacts. This application proposes a Retinex decoupling algorithm constrained by a saturated water absorption limit threshold, which rigidly introduces a "lower illuminance threshold determined by the physical reflectivity of the saturated fibers" when smoothing the illuminance component. This process forcibly decouples the low-frequency, large-area dark areas caused by moisture absorption from the high-spectral line absorption characteristics of carbon-based ink. After peeling off the wood substrate, local adaptive histogram equalization (CLAHE) is introduced only for the separated high-frequency carbon-based ink reflection components to control extreme value gain, thereby suppressing the noise of decayed wood grain, widening the local contrast of the ink stroke area, and finally outputting a high-definition digital orthophoto image of the bamboo and wooden slips with extremely sharp edges of the characters, that is, the enhanced image of the target bamboo and wooden slips.

[0093] In one feasible implementation, the step of performing local adaptive contrast enhancement processing on the high-frequency reflection component image of the unobstructed image to output the enhanced target bamboo slip image further includes steps E01 to E04: E01. Convert the unobstructed image to the logarithmic domain to obtain a logarithmic domain image; E02. Smooth the logarithmic domain image using a multi-scale Gaussian convolution function to obtain the low-frequency illumination component image. ; E03. Using the lower limit threshold of illuminance, the target illuminance value in the low-frequency illuminance component image that is lower than the lower limit threshold of illuminance is truncated and raised to the lower limit threshold of illuminance to generate a restricted illuminance component image. E04. Based on the limited illumination component image and the logarithmic domain image, the high-frequency reflection component image is decoupled.

[0094] That is, by extracting the low-frequency illumination component and the high-frequency reflection component of the unobstructed image, constraining the low-frequency illumination component with a pre-calibrated upper limit threshold, and performing local adaptive contrast enhancement on the high-frequency reflection component, the target bamboo slip image with enhanced ink edge is output.

[0095] In this application, "smoothing the illuminance component" refers to the calculation process of extracting the "illuminance component" that represents the slow-changing gradient between the ambient light over a large area and the wooden substrate by using a Gaussian filter to perform low-pass filtering on the image after removing scattered haze, within the framework of Retinex theory.

[0096] Specific processing flow: After step 204, we obtain a clean, unobstructed image that has "removed underwater suspended debris obstruction and completed physical inversion". .

[0097] Next, we proceed to the Retinex decoupling stage, steps E01 to E04. First, the image is transformed to the logarithmic domain, and then passed through a multi-scale Gaussian convolution function. The image is smoothed to estimate and extract low-frequency illuminance components. Furthermore, in this Gaussian smoothing estimation process, this application introduces a special hard constraint—the lower limit threshold Tmax of illuminance.

[0098] The physical source of the illuminance lower limit threshold Tmax: It is a constant lower limit threshold that is pre-calibrated and set based on the "physical reflectance lower limit benchmark of water-saturated wood fiber in the whole band".

[0099] In the conventional Retinex algorithm, the formula for calculating the illuminance component obtained by Gaussian smoothing is: = If a section of the wood substrate becomes extremely dark due to severe water absorption and decay, conventional Gaussian smoothing will faithfully record this extremely low local brightness as "illuminance." This results in the contrast of this extremely dark area being drastically amplified when the original image is subsequently divided by the illuminance to obtain the high-frequency reflection component, thus revealing the wood grain noise.

[0100] This application introduces Subsequently, when performing logarithmic field smoothing or component constraints, its mathematical expression is modified to: (4) This is equivalent to setting a "physical lower / upper limit protection lock for smoothed illumination." When the local illumination value calculated by Gaussian smoothing falls below this preset physical reference for saturated fibers... At that time, the algorithm forcibly "truncated" it and promoted it to In formula (4) L constrained (x,y) represents the low-frequency illuminance component.

[0101] In the low-frequency portion (wooden substrate): through this max constraint, the "extremely dark low-frequency background" originally caused by water absorption and resulting from large-area slow undulations is forcibly flattened and locked within... On the baseline, it no longer darkens infinitely with the degree of local decay.

[0102] High-frequency components (carbon-based ink): These are the extremely high-frequency, minute gray-level abrupt changes that truly represent the strokes of the characters and change spatially with incredible speed. Because their spatial frequency is much higher than the Gaussian filter window, they will not be affected. This process involves truncation, which forcibly and smoothly locks the low-frequency substrate while completely preserving the high-frequency ink marks.

[0103] Furthermore, through the above After successfully decoupling and separating the low-frequency water-saturated wood substrate from the high-frequency ink stains, a pure high-frequency reflection component matrix R(x,y) was obtained. To prevent the faint ink stains from remaining visually indistinguishable, the system then performed Contrast Limited Adaptive Histogram Equalization (CLAHE) for extreme gain control. The specific engineering process is as follows: Tile-based partitioning: The algorithm divides the entire high-frequency reflectance component image into several smaller, non-overlapping local rectangular grid tiles (e.g., 8 tiles). (8-pixel block).

[0104] Histogram Statistics and Clip Limit: This function calculates the grayscale probability distribution histogram of pixels within each local grid block. To prevent traditional histogram equalization from excessively amplifying noise (such as residual tiny wood fiber particles) and causing severe artifacts, CLAHE introduces a contrast clip limit. When the frequency of pixels at a certain grayscale level exceeds a preset threshold, the excess pixels are uniformly clipped and redistributed to other grayscale ranges, achieving "extreme gain control" (i.e., limiting the maximum magnification).

[0105] Cumulative Distribution Function (CDF) Calculation and Gray-Level Mapping: The cumulative distribution function (CDF) is calculated for the cropped local histogram to generate a gray-level nonlinear mapping curve specific to the grid block, stretching the gray-level difference between the faint ink blot and the background.

[0106] Bilinear Interpolation for Smooth Scraping: To eliminate the "block artifacts" that may be caused at grid boundaries by block processing, the algorithm uses bilinear interpolation to smooth weight fusion at the boundaries of adjacent grid blocks.

[0107] Final output: After this extreme gain control, the local contrast of the ink strokes is precisely broadened, while the background noise of the decayed wood grain is strictly suppressed. The final output is a high-definition orthophoto with extremely sharp edges of the characters. The target bamboo slip image includes this high-definition orthophoto.

[0108] The above-mentioned scanning device and image processing method have the following advantages and beneficial effects: First, it breaks through the physical limitations of dynamic refraction in traditional underwater photography, achieving high-precision geometric fidelity and high-throughput acquisition with a large field of view. This invention utilizes a structured optical path constraint of "flat-field waterproof cover + 400mm specific water depth + 8mm wide-angle lens" to transform the dynamic random refraction caused by water surface fluctuations into a statically reversible cross-medium offset, and combines this with a ray tracing model for precise compensation. This design not only achieves a global single exposure with a large field of view of approximately 280mm, significantly shortening the exposure time of waterlogged bamboo slips after removing them from their original protective liquid and increasing digitization throughput, but also ensures the strictly orthogonal geometric spatial accuracy of the unearthed bamboo slips' digitized archives from a physical source.

[0109] Secondly, it overcomes the failure problem of traditional single-image dehazing algorithms in dense bamboo and wooden slip scenarios, achieving precise separation of fog and suspended occlusion in slightly turbid water. This invention breaks the conventional underwater direct illumination pattern, creating an underwater grazing artificial light field strictly parallel to the surface of the bamboo and wooden slips, and combining it with a high-rigidity, lossless stepping rail to acquire multiview image sequences. This deep collaborative mechanism of "hardware multiview acquisition + software spatial coherence comparison" not only accurately inverts the spatial attenuation model of complex underwater artificial light field by extracting feature anchor points across frames, but also successfully removes the physical occlusion of macroscopic suspended particles in slightly turbid water by utilizing the minimum image patch mean intensity difference (LPMD), providing pure underlying data that traditional single-frame air scanning equipment cannot obtain for subsequent text extraction.

[0110] Third, it breaks through the technical bottleneck of spectral coupling between saturated wood and carbon-based ink, achieving high-contrast extraction of faint characters on the surface of decayed and blackened bamboo slips. Unlike conventional image enhancement algorithms applied to dry bamboo slips, this invention innovatively introduces a "saturated water absorption limit threshold" as a rigid constraint on the illuminance component in the Retinex model, specifically addressing the unique optical absorption degradation mechanism of the highly saturated wooden substrate. This mechanism forcibly decouples the blackened background of the wooden substrate from the high-frequency reflection components of the carbon-based ink, completely avoiding the erroneous amplification of damp wood grain noise by traditional multi-scale enhancement algorithms from a radiation physics perspective. This results in extremely sharp and high-fidelity rendering of the edges of ancient characters deeply embedded in the decayed material layer.

[0111] This invention provides an underwater image processing method for water-saturated bamboo slips. Through multi-view image acquisition combined with a layered processing architecture, it first addresses the cross-medium distortion problem in underwater imaging by constructing a ray tracing model based on Snell's law for nonlinear distortion correction. Simultaneously, it eliminates uneven illumination through three-dimensional flat-field normalization, resulting in a distortion-free, uniformly exposed pre-processed image. Secondly, based on an underwater imaging physical model, it constructs a global backscattering intensity vector using the darkest pixel information from the multi-view images, and removes the backscattering component pixel by pixel, effectively eliminating underwater glare and color distortion. Next, through multi-view image alignment and outlier removal and fusion, it eliminates transient occlusion interference from suspended particles, bubbles, etc., resulting in a clear, unobstructed base image. Finally, considering the material characteristics of the water-saturated bamboo slips, it constrains the low-frequency illuminance component using a pre-calibrated illuminance lower limit threshold based on the physical reflectivity lower limit of water-saturated wood fibers across the entire wavelength band. This decouples the high-frequency reflection component reflecting ink characteristics and performs local adaptive contrast enhancement, suppressing wood texture interference while strengthening the clarity of ink edges. This method eliminates the need for dehydration of waterlogged bamboo and wooden slips, thus avoiding damage to the artifacts. At the same time, the output images of the target bamboo and wooden slips have clear ink marks and low distortion, which can be directly used for text recognition and digital archiving. It supports batch and continuous processing, which greatly improves the efficiency and fidelity of digital acquisition of bamboo and wooden slips.

[0112] Please see Figure 3 , Figure 3 This is a structural block diagram of an underwater image processing device for water-saturated bamboo slips according to an embodiment of the present invention, as shown below. Figure 3 The apparatus shown includes: Image acquisition module 301: used to acquire an image sequence of waterlogged bamboo slips collected underwater, the image sequence including N frames of images with partial field of view overlap, the N frames of images having different viewpoints; Image preprocessing module 302: is used to perform cross-medium geometric distortion correction and three-dimensional flat field normalization preprocessing on each of the N frames of images in sequence to obtain the preprocessed N frames of images; Descattering module 303: Used to strip backscattering components from the preprocessed N frames of images by utilizing pixel information of corresponding spatial locations in the preprocessed N frames of images within the framework of an underwater imaging physical model, thereby obtaining N frames of descattered images; The occlusion removal module 304 is used to spatially align the N frames of descattered images, remove abnormal image blocks by comparing image blocks at the same spatial position and fuse and reconstruct them to obtain a single frame of occlusion-free image. Ink enhancement module 305: used to perform local adaptive contrast enhancement processing on the high-frequency reflection component image of the unobstructed image, and output the enhanced target bamboo slip image; the high-frequency reflection component image is used to reflect the ink characteristics of the water-saturated bamboo slip, the high-frequency reflection component image is obtained by decoupling from the limited illumination component image, the limited illumination component image is obtained by constraining the low-frequency illumination component image of the unobstructed image using a pre-calibrated lower illumination threshold, the lower illumination threshold is pre-calibrated based on the lower limit of physical reflectivity of water-saturated wood fibers in the whole band.

[0113] It should be noted that, Figure 3 The contents of each module in the device shown are... Figure 2 The steps in the method shown are similar, and will not be repeated here to avoid repetition. For details, please refer to [reference needed]. Figure 2 The content of each step in the method shown.

[0114] This invention provides an underwater image processing device for waterlogged bamboo slips. Through multi-view image acquisition combined with a layered processing architecture, it first addresses the cross-medium distortion problem in underwater imaging by constructing a ray tracing model based on Snell's law for nonlinear distortion correction. Simultaneously, it eliminates uneven illumination through three-dimensional flat-field normalization, resulting in a distortion-free, uniformly exposed pre-processed image. Secondly, based on an underwater imaging physical model, it constructs a global backscattering intensity vector using the darkest pixel information from the multi-view images, and removes the backscattering component pixel by pixel, effectively eliminating underwater glare and color distortion. Then, through multi-view image alignment and outlier removal and fusion, it eliminates transient occlusion interference from suspended particles, bubbles, etc., resulting in a clear, unobstructed base image. Finally, considering the material characteristics of the waterlogged bamboo slips, it constrains the low-frequency illuminance component using a pre-calibrated illuminance lower limit threshold based on the physical reflectivity lower limit of waterlogged wood fibers across the entire wavelength band. This decouples the high-frequency reflection component reflecting ink characteristics and performs local adaptive contrast enhancement, suppressing wood texture interference while strengthening the clarity of ink edges. This device eliminates the need for dehydration of waterlogged bamboo and wooden slips, thus avoiding damage to the artifacts. At the same time, the output images of the target bamboo and wooden slips have clear ink marks and low distortion, which can be directly used for text recognition and digital archiving. It supports batch continuous processing, which greatly improves the efficiency and fidelity of digital acquisition of bamboo and wooden slips.

[0115] Figure 4 An internal structural diagram of a computer device in one embodiment is shown. This computer device can specifically be a terminal or a server. Figure 4 As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and may also store a computer program, which, when executed by the processor, causes the processor to perform the aforementioned methods. The internal memory may also store a computer program, which, when executed by the processor, causes the processor to perform the aforementioned methods. Those skilled in the art will understand that… Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0116] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform actions such as... Figure 2 The steps of the method shown.

[0117] In one embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, causes the processor to perform the following actions: Figure 2 The steps of the method shown.

[0118] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0119] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0120] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for underwater image processing of a waterlogged document, characterized in that, The method includes: Acquire an image sequence of water-saturated bamboo slips collected underwater, the image sequence comprising N frames of images with partial field-of-view overlap, the N frames of images having different viewpoints; For each of the N frames, cross-medium geometric distortion correction and three-dimensional flat field normalization are performed sequentially to obtain the preprocessed N frames. Using the pixel information of the corresponding spatial locations in the preprocessed N frames of images, the backscattering component is removed from the preprocessed N frames of images within the framework of the underwater imaging physical model to obtain N frames of descattered images. Spatial alignment is performed on the N frames of descattered images. Abnormal image blocks are removed by comparing image blocks at the same spatial location and then fused and reconstructed to obtain a single frame of occlusion-free image. The high-frequency reflection component image of the unobstructed image is subjected to local adaptive contrast enhancement processing to output the enhanced target bamboo slip image. The high-frequency reflection component image is used to reflect the ink characteristics of the waterlogged bamboo slip. The high-frequency reflection component image is obtained by decoupling from the limited illumination component image. The limited illumination component image is obtained by constraining the low-frequency illumination component image of the unobstructed image using a pre-calibrated lower illumination threshold. The lower illumination threshold is pre-calibrated based on the lower limit of the physical reflectance of waterlogged wood fibers in the entire wavelength band.

2. The method of claim 1, wherein, The step of using pixel information at corresponding spatial locations in the preprocessed N frames of images to remove the backscattering component from the preprocessed N frames of images within the framework of an underwater imaging physical model, to obtain N frames of descattered images, includes: The preprocessed N frames of images are divided into M image blocks; The darkest pixel is extracted from each image block of each frame to form the backscatter intensity vector of each frame, resulting in N backscatter intensity vectors. By comparing the elements of the backscatter intensity vectors of different frames at corresponding spatial positions, the M global minimum values ​​of the M local regions corresponding to the M image blocks are determined. A multi-view backscattering intensity vector is constructed based on the M global minimum values; Based on the multi-view backscattering intensity vector, the coefficients of the artificial illumination distribution model are solved to generate a backscattering component image; Using predetermined water depth parameters, the backscattering component image, and a preset Jaffe-McGlamery model, scattering degradation inversion is performed on the N frames of geometrically corrected images. The backscattering component is stripped pixel by pixel from the N frames of geometrically corrected images to obtain N frames of descattered images. The underwater imaging physical model includes the Jaffe-McGlamery model.

3. The method of claim 1, wherein, If N is an odd number and N is greater than 1, then the spatial alignment of the N frames of descattered images includes: The middle frame of the N frames of descattered images is used as the reference frame; The feature points of each frame of the descattered image are obtained by using a preset feature detection algorithm to detect feature points in each frame. A preset feature point matching algorithm is used to perform feature point matching processing on the remaining frames other than the reference frame, so as to obtain matching feature point pairs between the remaining frames and the reference frame. Based on the feature point pairs, calculate the perspective transformation matrix of each of the remaining frames relative to the reference frame; Based on the perspective transformation matrix, the images of the remaining frames are transformed to the coordinate system of the reference frame to obtain a spatially aligned image.

4. The method of claim 3, wherein, The process of comparing and eliminating abnormal image blocks at the same spatial location and then fusing and reconstructing them to obtain a single frame of occlusion-free image includes: Extract the completely overlapping common viewing regions from the spatially aligned image and determine them as regions of interest; The region of interest is divided into multiple image blocks; The average intensity difference is calculated by pairing image patches at the same spatial location; Abnormal image blocks with average intensity differences exceeding a preset threshold are removed; and the image block with the smallest average intensity difference is selected and its pixel-wise average is calculated for fusion and reconstruction to obtain a single frame of unobstructed image.

5. The method of claim 1, wherein, The step of performing local adaptive contrast enhancement processing on the high-frequency reflection component image of the unobstructed image to output the enhanced target bamboo slip image also includes: The unobstructed image is converted to the logarithmic domain to obtain a logarithmic domain image; The logarithmic domain image is smoothed by a multi-scale Gaussian convolution function to obtain a low-frequency illumination component image; Using the lower limit threshold, the target illuminance values ​​in the low-frequency illuminance component image that are below the lower limit threshold are truncated and raised to the lower limit threshold to generate a restricted illuminance component image. Based on the limited illumination component image and the logarithmic domain image, the high-frequency reflection component image is decoupled.

6. The method according to claim 1, characterized in that, For each of the N frames, preprocessing is performed sequentially, including cross-medium geometric distortion correction and three-dimensional flat-field normalization, to obtain the preprocessed N frames, including: The three-dimensional flat-field normalization algorithm, the white field image and the black field image of different color channels are pre-calibrated to perform three-dimensional flat-field normalization processing on each frame of the image pixel by pixel and color channel by color channel to obtain N frames of normalized images. A cross-medium ray tracing model is constructed based on Snell's law. The thickness and refractive index of the waterproof cover glass, the refractive index of the preservation liquid, and the vertical object distance from the camera optical center to the surface of the water-saturated bamboo slip are used as known constant inputs. The ray path from the camera pixel coordinates through each medium layer to the surface of the water-saturated bamboo slip is traced in reverse. The nonlinear offset vector of each pixel in the normalized image of each frame is calculated. Based on the offset vector, spatial coordinate remapping is performed on the normalized images of each frame to obtain a corrected image that eliminates cross-medium distortion. The preprocessed N-frame images include the corrected image.

7. An underwater image processing device for water-saturated bamboo and wooden slips, characterized in that, The device includes: Image acquisition module: used to acquire an image sequence of waterlogged bamboo slips collected underwater, the image sequence including N frames of images with partial field of view overlap, the N frames of images having different viewpoints; Image preprocessing module: used to sequentially perform cross-medium geometric distortion correction and three-dimensional flat field normalization preprocessing on each of the N frames of images to obtain the preprocessed N frames of images; Descattering module: Used to extract the backscattering component from the preprocessed N frames of images by utilizing the pixel information of the corresponding spatial positions in the preprocessed N frames of images within the framework of the underwater imaging physical model, so as to obtain N frames of descattered images; Occlusion removal module: used to spatially align the N frames of descattered images, remove abnormal image blocks by comparing image blocks at the same spatial position and fuse and reconstruct them to obtain a single frame of occlusion-free image; Ink Enhancement Module: Used to perform local adaptive contrast enhancement processing on the high-frequency reflection component image of the unobstructed image, and output the enhanced target bamboo slip image; the high-frequency reflection component image is used to reflect the ink characteristics of the waterlogged bamboo slip, and the high-frequency reflection component image is obtained by decoupling from the limited illumination component image. The limited illumination component image is obtained by constraining the low-frequency illumination component image of the unobstructed image using a pre-calibrated lower illumination threshold. The lower illumination threshold is pre-calibrated based on the lower limit of the physical reflectance of waterlogged wood fibers in the entire wavelength band.

8. An underwater multi-view scanning device for bamboo and wooden slips, characterized in that, The underwater bamboo and wooden slips multi-view scanning device includes: A water tank, the bottom of which is used to hold several bamboo slips, and the water tank is used to provide an underwater environment for collecting images of the bamboo slips; A feed actuator is mounted on top of the water tank; An image acquisition device is mounted on the feed execution component and moves with the feed execution component to acquire images of the bamboo slips at the bottom of the water tank during the feeding process, so as to obtain an image sequence of the water-saturated bamboo slips. A flat-field waterproof cover assembly is fitted over the outside of the image acquisition device. During image acquisition, the bottom of the flat-field waterproof cover assembly extends below or is close to the liquid surface of the water tank. A ring-shaped shadowless soft light illumination array surrounds the water level line attached to the inner wall of the water tank, providing a shadowless environment during image acquisition. The main control unit is connected to the ring-shaped shadowless soft light illumination array, the image acquisition device, and the feed execution component to send working instructions to the ring-shaped shadowless soft light illumination array, the image acquisition device, and the feed execution component. The main control unit is also used to perform the steps of the method as described in any one of claims 1-6.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it causes the processor to perform the steps of the method as described in any one of claims 1 to 6.

10. A computer device, comprising a memory and a processor, characterized in that, The memory stores a computer program that, when executed by the processor, causes the processor to perform the steps of the method as described in any one of claims 1 to 6.

Images