Image dataset generation system and method for computed imaging

Abstract

This disclosure relates to an image dataset generation system and method for computational imaging. The system includes: a display device for displaying one or more frames of ground truth images; an image acquisition device for acquiring images from the ground truth images to obtain sample degraded images, wherein the imaging metrics of the sample degraded images are lower than those of the corresponding ground truth images; and a control device for generating an image dataset based on the ground truth images and sample degraded images to train a computational imaging model, enabling the trained computational imaging model to process the degraded images to be processed, thereby obtaining enhanced images. According to embodiments of this disclosure, a pairwise image dataset with pixel-level correspondence between "ground truth images and sample degraded images" can be generated under conditions of low equipment cost and high acquisition efficiency. This dataset combines content diversity and physical realism, making it suitable for the training and evaluation of computational imaging models, thus providing stable and scalable data support for the development of data-driven computational imaging algorithms.

Description

Technical Field

[0001] This disclosure relates to the field of computational imaging technology, and in particular to an image dataset generation system and method for computational imaging. Background Technology

[0002] With the continuous development of semiconductor technology and optoelectronic technology, digital imaging equipment has been widely used in scientific research, industrial inspection, and everyday consumer electronics. However, especially in consumer electronics products such as mobile phones, the physical hardware components of the imaging system (such as optical lenses and image sensors) face objective bottlenecks in performance improvement due to factors such as device size, manufacturing cost, and power consumption. It is difficult to continuously improve image quality solely through hardware means. Against this backdrop, computational imaging has emerged. It no longer treats optical hardware and post-processing software as independent parts, but rather designs them together, using powerful computing capabilities to compensate for or even surpass the imaging limitations of pure physical optics.

[0003] In the actual imaging process, image quality degradation stems from multiple physical factors. Firstly, the optical system itself introduces image quality loss: due to the dispersive properties and geometric limitations of lens materials, aberrations occur, such as chromatic aberration preventing different colors of light from converging at the same point, and spherical aberration and coma causing uneven sharpness between the center and edges of the image. Furthermore, geometric distortion and vignetting also exist. Secondly, image sensors introduce information loss and noise during photoelectric conversion: the randomness of photons reaching pixels leads to shot noise; the sensor circuitry itself introduces readout noise and thermal noise; the Bayer color filter array covering the sensor allows each pixel to record only one color information, making it highly susceptible to artifacts (such as moiré patterns and zipper effects) when recovering a full-color image using demosaicing algorithms.

[0004] To address the aforementioned imaging degradation problem, various computational imaging schemes have been proposed in existing technologies. Among them, data-driven computational imaging methods have gradually gained attention. These methods learn the mapping relationship between degraded observations and high-quality images from a large amount of sample data by constructing algorithmic models to improve image quality. However, the performance of current data-driven computational imaging schemes largely depends on the size and quality of the training dataset. Therefore, how to efficiently and cost-effectively acquire a training dataset that accurately reflects the real physical degradation process and possesses high-quality ground-value images has become a key bottleneck restricting the development of computational imaging technology as a whole. Summary of the Invention

[0005] In view of this, this disclosure proposes an image dataset generation system and method for computational imaging.

[0006] According to one aspect of this disclosure, an image dataset generation system for computational imaging is provided. The system includes:

[0007] Display devices are used to display one or more frames of truth images;

[0008] An image acquisition device is used to acquire images from a ground truth image to obtain a sample degraded image that corresponds one-to-one with each frame of the ground truth image. The imaging parameters of the sample degraded image are lower than those of the corresponding ground truth image. The imaging parameters include at least one of resolution, brightness, color, and signal-to-noise ratio.

[0009] The control device, which is connected to the display device and the image acquisition device respectively, is used to obtain an image dataset based on the ground truth image and the sample degraded image. The image dataset is used to train a computational imaging model, so that the trained computational imaging model can process the degraded image to be processed acquired by the image acquisition device to obtain an enhanced image. The imaging index of the enhanced image is higher than that of the degraded image to be processed.

[0010] In one possible implementation, the system further includes a mechanical displacement device configured to support the display device and / or image acquisition device for adjusting the relative positions of the display device and the image acquisition device.

[0011] In one possible implementation, the mechanical displacement device includes a robotic arm, which controls the movement of the image acquisition device according to preset motion parameters when the image acquisition device is acquiring images. The preset motion parameters include one or more of motion trajectory, motion speed, acceleration, and vibration frequency.

[0012] In one possible implementation, the system further includes an optical degradation simulation device placed between the display device and the image acquisition device, for reducing any one or more imaging parameters of the sample degraded image relative to the ground truth image.

[0013] In one possible implementation, the optical degradation simulation device includes a neutral density filter used to attenuate the luminous flux passing through the imaging optical path, so that the brightness and / or signal-to-noise ratio of the sample degradation image is lower than that of the true image.

[0014] In one possible implementation, the optical degradation simulation device includes a medium simulation chamber for introducing spectrally selective absorption and / or scattering effects into the imaging optical path to reduce the contrast and / or color shift of the degraded image of the sample to that of the true image.

[0015] In one possible implementation, the system also includes a light-shielding dark box, which is placed outside the display device and the image acquisition device and at least covers the imaging optical path area between the display device and the image acquisition device, in order to isolate external ambient light interference or simulate ambient lighting conditions when the image acquisition device is acquiring images.

[0016] In one possible implementation, the control device is used for:

[0017] Based on the mapping relationship between the distorted coordinates of pixels in the degraded sample image and the corresponding pixel coordinates in the ground truth image, coordinate correction is performed on the degraded sample image to obtain a corrected degraded sample image aligned with the ground truth image.

[0018] The image dataset is obtained based on the ground truth image and the corrected sample degraded image.

[0019] In one possible implementation, the control device is used for:

[0020] Without placing an optical degradation simulation device, the display device and image acquisition device are controlled to acquire a first image dataset;

[0021] With the optical degradation simulation device in place, the display device and image acquisition device are controlled to acquire a second image dataset;

[0022] Training the computational imaging model includes:

[0023] The initial computational imaging model was pre-trained using the first image dataset to obtain the pre-trained computational imaging model.

[0024] The pre-trained computational imaging model was fine-tuned using the second image dataset to obtain the fine-tuned computational imaging model.

[0025] Among them, the ground truth images in the second image dataset are related to the target imaging scene targeted by the fine-tuning training, and the imaging indicators adjusted by the optical degradation simulation device are related to the target imaging scene.

[0026] According to another aspect of this disclosure, a method for generating image datasets for computational imaging is provided. The method includes:

[0027] The display device displays one or more frames of truth images;

[0028] The image acquisition device acquires images from the ground truth image to obtain a sample degraded image that corresponds one-to-one with each frame of the ground truth image. The imaging index of the sample degraded image is lower than that of the corresponding ground truth image. The imaging index includes at least one of resolution, brightness, color, and signal-to-noise ratio.

[0029] The control devices connected to the display device and the image acquisition device respectively obtain an image dataset based on the ground truth image and the sample degraded image. The image dataset is used to train the computational imaging model, so that the trained computational imaging model can process the degraded image to be processed acquired by the image acquisition device to obtain an enhanced image. The imaging index of the enhanced image is higher than that of the degraded image to be processed.

[0030] In one possible implementation, an image dataset is obtained based on the ground truth image and the degraded sample images, including:

[0031] Based on the mapping relationship between the distorted coordinates of pixels in the degraded sample image and the corresponding pixel coordinates in the ground truth image, coordinate correction is performed on the degraded sample image to obtain a corrected degraded sample image aligned with the ground truth image.

[0032] The image dataset is obtained based on the ground truth image and the corrected sample degraded image.

[0033] In one possible implementation, the method further includes:

[0034] Without placing an optical degradation simulation device, the control device controls the display device and the image acquisition device to acquire a first image dataset;

[0035] With the optical degradation simulation device in place, the control equipment controls the display device and the image acquisition device to acquire a second image dataset;

[0036] Training the computational imaging model includes:

[0037] The initial computational imaging model was pre-trained using the first image dataset to obtain the pre-trained computational imaging model.

[0038] The pre-trained computational imaging model was fine-tuned using the second image dataset to obtain the fine-tuned computational imaging model.

[0039] Among them, the ground truth images in the second image dataset are related to the target imaging scene targeted by the fine-tuning training, and the imaging indicators adjusted by the optical degradation simulation device are related to the target imaging scene.

[0040] According to another aspect of this disclosure, an image dataset generation apparatus for computational imaging is provided, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above-described method.

[0041] According to another aspect of this disclosure, a non-volatile computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the steps of the above-described method.

[0042] According to another aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps of the above-described method.

[0043] According to embodiments of this disclosure, the ground truth image output by the display device can be used as a controllable, lossless source of truth. Corresponding degraded observation results (i.e., sample degraded images) acquired by the image acquisition device can be obtained through a real physical imaging link. This introduces the physical degradation effects generated by the real optical system and image sensor without relying on complex simulation modeling. This ensures that the ground truth image content is flexibly configurable while the generated sample degraded images accurately reflect the imaging characteristics of the target imaging device (i.e., the image acquisition device) under actual working conditions. Embodiments of this disclosure can generate a pairwise image dataset of "ground truth image - sample degraded image" with pixel-level correspondence under conditions of low equipment cost and high acquisition efficiency. This image dataset combines content diversity and physical realism, making it suitable for training and evaluating computational imaging models, thereby providing stable and scalable data support for the development of data-driven computational imaging algorithms.

[0044] Other features and aspects of this disclosure will become clear from the following detailed description of exemplary embodiments with reference to the accompanying drawings. Attached Figure Description

[0045] The accompanying drawings, which are included in and form part of this specification, illustrate exemplary embodiments, features, and aspects of this disclosure together with the specification and serve to explain the principles of this disclosure.

[0046] Figure 1 A structural diagram of an image dataset generation system for computational imaging according to an embodiment of the present disclosure is shown.

[0047] Figure 2 A schematic diagram of an image dataset generation system for computational imaging according to an embodiment of the present disclosure is shown.

[0048] Figure 3 A schematic diagram of an image dataset generation system for computational imaging according to an embodiment of the present disclosure is shown.

[0049] Figure 4 A schematic diagram illustrating the training and testing phases of a computational imaging model according to an embodiment of the present disclosure is shown.

[0050] Figure 5 A flowchart is shown for a method of generating an image dataset for computational imaging according to an embodiment of the present disclosure. Detailed Implementation

[0051] Various exemplary embodiments, features, and aspects of this disclosure will now be described in detail with reference to the accompanying drawings. The same reference numerals in the drawings denote elements that have the same or similar functions. Although various aspects of the embodiments are shown in the drawings, they are not necessarily drawn to scale unless specifically indicated otherwise.

[0052] As used herein, the terms “comprising,” “including,” “having,” or variations thereof are open-ended and include one or more of the stated features, integrals, elements, steps, components, or functions, but do not exclude the presence or addition of one or more other features, integrals, elements, steps, components, functions, or groups thereof.

[0053] When an element is referred to as “connected,” “coupled,” “responding,” or a variation thereof relative to another element, it may be directly connected, coupled, or responding to another element, or there may be an intermediate element present.

[0054] Although the terms first, second, third, etc., may be used herein to describe various elements / operations, these elements / operations should not be limited by these terms. These terms are only used to distinguish one element / operation from another. Therefore, without departing from the teachings of the inventive concept, a first element / operation in some embodiments may be referred to as a second element / operation in other embodiments.

[0055] The term “exemplary” as used herein means “serving as an example, embodiment, or illustration.” Any embodiment illustrated herein as “exemplary” is not necessarily to be construed as superior to or better than other embodiments.

[0056] Furthermore, to better illustrate this disclosure, numerous specific details are set forth in the following detailed description. Those skilled in the art will understand that this disclosure can be practiced without certain specific details. In some instances, methods, means, components, and circuits well known to those skilled in the art have not been described in detail in order to highlight the main points of this disclosure.

[0057] It should be noted that the information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, data stored, data displayed, etc.) and signals involved in this application are all authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant regions.

[0058] In the current field of computational imaging, especially in the development of data-driven computational imaging methods, the acquisition of training datasets still faces two major challenges:

[0059] 1. There is a mismatch between simulation-based datasets and real physical imaging processes. While pure software simulation offers advantages in terms of data generation flexibility and cost control, it typically relies on simplified physical models, making it difficult to comprehensively and accurately characterize the complex imaging degradation in real optical systems. For example, aberrations, chromatic aberration, and assembly errors commonly found in actual optical lenses are difficult to fully reproduce in simulation models. Furthermore, the inherent noise characteristics, nonlinear responses, and inter-pixel crosstalk of image sensors during operation are also difficult to accurately describe using idealized models. Therefore, computational imaging models trained on such simulation data often exhibit performance degradation or insufficient generalization ability when applied to real physical imaging devices.

[0060] 2. Training data based on real-world scene acquisition also faces limitations. Existing technologies often employ a "high-end reference camera" and a "target camera" to simultaneously acquire data from the same real-world scene in order to obtain paired training samples. While this approach ensures that the data originates from the actual physical imaging process, it typically requires high-cost imaging equipment and complex synchronization and calibration processes, resulting in low data acquisition efficiency and limitations in scaling the scene. More importantly, due to differences in optical structure, imaging geometry, and sampling methods among different imaging systems, this type of method struggles to obtain an ideal ground truth image that is precisely aligned with the target camera's image at the pixel level without any information loss, thus constraining the quality of the training data.

[0061] In view of this, this disclosure provides an image dataset generation system and method for computational imaging. The system of this disclosure can utilize the ground truth image output by a display device as a controllable, lossless source of truth, and acquire the corresponding degraded observation results (i.e., sample degraded images) acquired by the image acquisition device through a real physical imaging link. This introduces the physical degradation effects generated by the real optical system and image sensor without relying on complex simulation modeling, ensuring that the ground truth image content is flexibly configurable while the generated sample degraded images accurately reflect the imaging characteristics of the target imaging device (i.e., the image acquisition device) under actual working conditions. This disclosure can generate a pairwise image dataset of "ground truth image - sample degraded image" with pixel-level correspondence under conditions of low equipment cost and high acquisition efficiency. This image dataset combines content diversity and physical realism, making it suitable for training and evaluating computational imaging models, thereby providing stable and scalable data support for the development of data-driven computational imaging algorithms.

[0062] Figure 1 A structural diagram of an image dataset generation system for computational imaging according to an embodiment of the present disclosure is shown. Figure 1As shown, the system includes:

[0063] Display device 100 is used to display one or more frames of true image;

[0064] Image acquisition device 200 is used to acquire images of ground truth images and obtain sample degraded images that correspond one-to-one with each frame of ground truth image;

[0065] The control device 300, which is connected to the display device 100 and the image acquisition device 200 respectively, is used to obtain an image dataset based on the ground truth image and the sample degraded image.

[0066] The display device 100 can be any type of display terminal, such as a liquid crystal display (LCD), an organic light-emitting diode display (OLED), a microdisplay, or other display devices capable of digitally presenting image content. The function of the display device 100 is to display the true value image as the "scene truth value" with high fidelity, thereby providing a controllable, repeatable, and complete standard reference input for the subsequent image acquisition process.

[0067] In computational imaging scenarios, the display device 100 can be a professional-grade display device with high resolution, high color accuracy, and high refresh rate to ensure that the displayed image has sufficient fidelity in terms of spatial resolution, brightness level, color distribution, and grayscale details.

[0068] Optionally, the display device 100 can also be a display array, which may consist of multiple high-brightness, ultra-high-contrast panels. The display device 100 may also employ a display structure with local backlight control capabilities, achieving enhanced display of high-brightness areas by finely adjusting the brightness of different regions. Through the aforementioned display array or local backlight enhancement method, the display device 100 can simulate high dynamic range lighting conditions in natural scenes, such as backlit scenes, scenes with strong direct light sources, or complex lighting environments with significant contrast between light and dark. Compared to conventional display devices, which are limited by peak brightness and overall backlight structure and thus struggle to reproduce high-brightness details, the above method can provide a brightness distribution at the display end that more closely resembles the real physical scene.

[0069] Therefore, when the image acquisition device captures the displayed content, an imaging degradation process approaching overexposure or high-brightness saturation conditions can be introduced into the real physical imaging chain, thereby obtaining sample degradation images with clear physical meaning. The generated image dataset can support the training and validation of computational imaging models for scenarios such as high dynamic range reconstruction and overexposure detail recovery.

[0070] A ground truth image can refer to image data that has been generated or acquired in advance. For example, a ground truth image can be a digital image generated by computer software, or an image acquired by a high-end reference camera under near-ideal imaging conditions. Under ideal imaging conditions, a ground truth image represents the true content of the target imaging scene and is unaffected by the optical system, sensor noise, or physical imaging link degradation factors of the image acquisition device 200, or the degree of influence of these factors is negligible.

[0071] Image acquisition device 200 can be a digital camera, industrial camera, mobile terminal camera module, or any other type of imaging device with image acquisition capabilities. In computational imaging applications, image acquisition device 200 can be a target camera to be studied or optimized, acquiring images of a specific target imaging scene to obtain degraded images reflecting the true imaging characteristics of the target camera. Image acquisition device 200 can consist of an optical lens under test and an image sensor, forming a complete physical imaging link that needs to be modeled or compensated.

[0072] The physical imaging link is used to characterize the imaging degradation factors introduced by optical systems, photoelectric conversion, and signal acquisition during the real imaging process. These degradation factors include, but are not limited to, optical aberrations, dispersion effects, image sensor noise, nonlinear response, and inter-pixel crosstalk. By modeling and learning this physical imaging link, the subsequently trained computational imaging model can be specifically optimized for the imaging characteristics of the image acquisition device 200.

[0073] Because the physical imaging link introduces the aforementioned imaging degradation factors during the actual imaging process, the quality of the sample degraded image acquired by the image acquisition device 200 will be lower than that of the true image displayed by the display device 100. Specifically, the imaging parameters of the sample degraded image can be lower than those of the corresponding true image. Imaging parameters can include at least one of resolution, brightness, color, and signal-to-noise ratio. A higher color imaging parameter indicates a more accurate correspondence between the RGB color values ​​in the sample degraded image and the true image, meaning less proportional distortion in each RGB channel and less color shift and saturation difference. Conversely, a lower color imaging parameter indicates a less accurate correspondence between the RGB color values ​​in the sample degraded image and the true image, meaning greater proportional distortion in each RGB channel and greater color shift and saturation difference.

[0074] This disclosure employs an image acquisition path based on a real physical imaging link. Although the ground truth image can exist digitally and be displayed by the display device 100, the imaging degradation process from the ground truth image to the sample degradation image is completed at the physical level by the image acquisition device 200, which includes a real optical lens and an image sensor. Compared to pure software simulation methods, this approach inherently incorporates various real physical effects that are difficult to describe precisely through mathematical models, thereby significantly improving the physical authenticity and engineering applicability of the "ground truth - measured value" image pairs in the generated image dataset.

[0075] The control device 300 can be a terminal device or a server. The terminal device can be any one or more of the following: mobile phone, foldable electronic device, tablet computer, desktop computer, laptop computer, handheld computer, notebook computer, ultra-mobile personal computer (UMPC), netbook, cellular phone, personal digital assistant (PDA), or in-vehicle device, with wired or wireless communication capabilities. This disclosure does not impose any special restrictions on the specific type of terminal device. The server can be located locally or in the cloud, and can be a physical device or a virtual device, such as a virtual machine or container, with wireless communication capabilities. These wireless communication capabilities can be configured in the server's chip (system) or other components. Wireless communication capabilities can be implemented, for example, through 2G / 3G / 4G / 5G mobile communication technologies, as well as Wi-Fi, Bluetooth, frequency modulation (FM), data radio, satellite communication, etc.; or through wired connections to enable interaction with other devices.

[0076] The control device 300 can also be used to coordinate the control of the display device 100 and the image acquisition device 200. Specifically, the control device 300 can control the display parameters of the display device 100, including but not limited to display resolution, refresh rate, brightness, contrast ratio, color mode, and backlight driving method; the control device 300 can also control the image acquisition parameters of the image acquisition device 200, including but not limited to exposure time, gain, sensitivity, white balance mode, shutter speed, and frame rate. Optionally, the control device 300 can combine the light emission characteristics of the display device 100, or adjust the exposure time and integration method of the image acquisition device 200, to match the image acquisition process with the refresh rate of the display device 100, thereby reducing or eliminating flicker artifacts introduced by display refresh or pulsed light emission.

[0077] During image acquisition, the control device 300 can obtain the first ground truth image from a preset standard digital image library and send it to the display device 100 for display (e.g., full-screen display). Simultaneously, the control device 300 can also control the image acquisition device 200 to capture images of the content displayed on the display device 100 according to the preset image acquisition parameters via an acquisition control program, thereby obtaining a sample degradation image corresponding to the ground truth image. The control device 300 can receive the sample degradation image sent by the image acquisition device 200, pair and associate the sample degradation image with the corresponding ground truth image, and store it in a storage device. Then, the control device 300 can continue to acquire the next ground truth image and repeat the above process until the predetermined size of the image dataset is acquired.

[0078] Optionally, the control device 300 can also acquire the video source file and parse it into a continuous sequence of frame images. In this case, the control device 300 can control the display device 100 to display the frame image sequence frame by frame according to the video frame order, and synchronously trigger the image acquisition device 200 to acquire images frame by frame. The control device 300 can align and associate the acquired continuous sample degraded image sequence with the original video frame sequence according to the frame order to construct a video data pair containing temporal domain information, as an image dataset.

[0079] Through the above methods, the system of this embodiment, while maintaining the source of the ground truth image content as a digital image, ensures that the degradation process from the ground truth image to the sample degraded image is completed entirely through the real physical imaging link, thereby generating an image dataset that simultaneously possesses content diversity and physical realism. This dataset can accurately reflect the imaging degradation characteristics of the target image acquisition device 200 under real imaging conditions. Furthermore, the system of this embodiment constructs an acquisition environment based on the display device 100 and the target image acquisition device 200, allowing for acquisition in a laboratory setting at a lower cost than real-world acquisition, thus reducing data acquisition costs and simplifying the acquisition process. By replacing or updating the input digital ground truth image source, the size of the image dataset can be flexibly expanded without changing the physical imaging link structure to meet the training or optimization needs of computational imaging models in different imaging scenarios (e.g., low illumination, high dynamic range, etc.).

[0080] In this embodiment, the ground truth image can be used as a reference image in the training data to form a one-to-one "ground truth - measured value" image pair with the sample degraded images acquired by the image acquisition device 200, thus obtaining an image dataset. The image dataset can be used to train a computational imaging model, enabling the trained model to process the degraded images acquired by the image acquisition device 200 to obtain enhanced images. The enhanced images have higher imaging metrics than the degraded images, and these metrics are associated with the training objective. During training, the sample degraded images are used as input to the computational imaging model, and the ground truth images are used as the training objective, i.e., the label data. The training process will be described later.

[0081] In one possible implementation, the system further includes an optical degradation simulation device 500, which can be placed between the display device 100 and the image acquisition device 200, for example, in the imaging optical path of the image acquisition device 200, to introduce controllable optical degradation when the true image is acquired via a physical imaging process, so that any one or more imaging indices of the sample degraded image are reduced relative to the true image.

[0082] The type of optical degradation simulation device 500 can be determined based on needs and associated with the training target of the subsequent computational imaging model, such as a neutral density filter, a medium simulation chamber, a scattering plate, etc., to simulate the degradation process under specific imaging scenarios or imaging conditions at the physical level.

[0083] By introducing the optical degradation simulation device 500, the true image will pass through the optical degradation simulation device 500, the optical lens of the image acquisition device 200 and the image sensor in sequence before being acquired by the image acquisition device 200, thus forming a complete physical imaging link.

[0084] Therefore, the degraded sample images are not digitally synthesized, but rather are raw sensor data acquired by the image acquisition device 200 after the ground truth image has been degraded through the aforementioned complete physical imaging chain. The ground truth image displayed on the display device 100 and the corresponding acquired degraded sample images constitute a set of training data pairs that are one-to-one corresponding in content and interpretable and reproducible along the physical degradation path. This solves the problems of data generation realism and determinism, making it more suitable for hardware-in-the-loop simulation and evaluation requiring high-precision physical models.

[0085] In one possible implementation, the optical degradation simulation device 500 may include a neutral density filter, which can be used to attenuate the luminous flux passing through the imaging optical path without changing the spectral composition distribution, so that the brightness and / or signal-to-noise ratio of the sample degradation image is lower than that of the true image.

[0086] A neutral density filter can be placed in front of the lens of the image acquisition device 200 to simulate specific shooting scenarios such as low light and night scenes.

[0087] In one possible implementation, the optical degradation simulation device 500 may include a medium simulation chamber for introducing spectrally selective absorption and / or scattering effects into the imaging optical path to make the contrast and / or color shift of the sample degradation image lower than that of the true image.

[0088] The medium simulation chamber can be sealed and transparent. By pre-filling the chamber with different concentrations of smoke, water vapor, or chemically suspended media, a parameter-controllable scattering medium environment can be constructed to simulate imaging degradation caused by medium scattering in scenarios such as fog, haze, or underwater imaging. The resulting image dataset can realistically reflect the physical transport process of photons in the scattering medium. This image dataset can support the training of computational imaging models for scenarios such as defogging, noise reduction, and transmission scattering.

[0089] In one possible implementation, the system may further include a light-shielding dark box 600, which may be placed outside the display device 100 and the image acquisition device 200, and at least cover the imaging optical path area between the display device 100 and the image acquisition device 200, for isolating external ambient light interference or simulating ambient lighting conditions when the image acquisition device 200 is acquiring images.

[0090] When the darkroom 600 is used to isolate external ambient light interference during image acquisition by the image acquisition device 200, it can be configured as an opaque, sealed structure. Its inner walls can be treated with low-reflection or anti-reflection coatings to suppress the generation of stray reflected light. By providing an acquisition environment similar to a darkroom, interference from ambient stray light on the acquisition process can be effectively eliminated, ensuring that the light signal received by the image acquisition device 200 originates entirely from the display content of the display device 100.

[0091] In the case of the darkroom 600 used to simulate ambient lighting conditions during image acquisition by the image acquisition device 200, a controllable auxiliary light source system can also be installed inside. This auxiliary light source system, for example, consists of a micro LED array, and its light source parameters, such as the position, emission direction, color temperature, light intensity, and flicker frequency, are controlled by the control device 300. By setting the above light source parameters, various complex lighting environments can be simulated at the physical level, such as the glare effect caused by backlighting, the changes in brightness and darkness caused by side lighting, and the periodic flicker effect introduced by artificial light sources. Therefore, the acquired sample degraded images can contain realistic optical stray light and complex tonal degradation, effectively providing training data for subsequent computational imaging algorithms such as high dynamic range imaging, glare removal, and color constancy.

[0092] In one possible implementation, the system may further include a mechanical displacement device 400 configured to support the display device 100 and / or the image acquisition device 200, for adjusting the relative positions of the display device 100 and the image acquisition device 200.

[0093] The mechanical displacement device 400 can be any type of controllable displacement mechanism, such as a linear guide rail, a two-dimensional or three-dimensional displacement platform, an optical platform, or a combination thereof, to achieve precise adjustment of the spatial position of the display device 100 and the image acquisition device 200. The relative position can at least include the relative distance and relative angle between the display device 100 and the image acquisition device 200, to meet the requirements of system geometric calibration, imaging field of view matching, and spatial resolution matching.

[0094] Optionally, considering the discretization characteristics of the content displayed by the display device 100, the theoretical sampling resolution of the image acquisition device 200 under the current imaging configuration can be calculated based on the diffraction limit of the optical system and the Nyquist sampling conditions of the sensor of the image acquisition device 200. The mechanical displacement device 400 can be adjusted so that the distance between the image acquisition device 200 and the display device 100 is such that the sampling resolution of the image acquisition device 200 is less than a preset multiple (e.g., twice) of the display resolution of the display device 100, thereby suppressing or avoiding moiré fringe artifacts caused by sampling mismatch at the physical imaging optical path level.

[0095] Figure 2 A schematic diagram of an image dataset generation system for computational imaging according to an embodiment of the present disclosure is shown. When the mechanical displacement device 400 is implemented in the form of a linear guide rail, the image dataset generation system for computational imaging according to an embodiment of the present disclosure is as follows: Figure 2 As shown.

[0096] In one possible implementation, the mechanical displacement device 400 may include a robotic arm, which controls the movement of the image acquisition device 200 according to preset motion parameters when the image acquisition device 200 is acquiring images. The preset motion parameters include one or more of motion trajectory, motion speed, acceleration and vibration frequency.

[0097] This robotic arm can be any type of actuator with multi-degree-of-freedom motion control capabilities, such as a six-axis industrial robotic arm. By setting motion parameters, the robotic arm can simulate random shaking during handheld imaging, dynamic blur caused by the relative motion of the imaging subject or imaging device, and dynamic imaging changes introduced by relative motion during zooming or focusing. The sample degraded images obtained based on the above method can contain imaging degradation features introduced by controlled motion, thereby constructing an image dataset containing specific motion degradation factors. This can support the training or validation of computational imaging models for electronic image stabilization, video deblurring, and dynamic scene imaging.

[0098] Figure 3 A schematic diagram of an image dataset generation system for computational imaging according to an embodiment of the present disclosure is shown. Figure 3 As shown, when the mechanical displacement device 400 is implemented in the form of a robotic arm, the robotic arm can be used to support and drive the image acquisition device 200.

[0099] Because geometric distortions, such as barrel distortion or pincushion distortion, are inevitably introduced during the imaging process in the embodiments of this disclosure, the acquired sample degraded image may be nonlinearly distorted in the spatial geometry, thus failing to achieve precise alignment with the true image displayed by the display device 100 at the pixel level. This geometric mismatch is particularly pronounced in the edge region of the imaging field of view.

[0100] To solve the above problem, a calibration image (such as a checkerboard image, a dot matrix image, or other calibration image with a known geometric structure) can be pre-displayed on the display device 100. The calibration image is then acquired by an image acquisition device to obtain a distorted image containing geometric distortion, and the distorted image is sent to the control device 300.

[0101] The control device 300 can extract feature points from the distorted image and the corresponding calibration image. Feature point extraction methods can include corner detection, edge detection, or template matching. The extracted feature point coordinates can be refined through interpolation, fitting, or other methods, or the feature correspondence can be determined using a dense matching algorithm based on optical flow. Based on the feature point matching results, the control device 300 can calculate distortion correction parameters to describe the geometric distortion. These distortion correction parameters can be a mapping relationship between the distorted coordinates of pixels in the distorted image and the corresponding pixel coordinates in the calibration image.

[0102] In one possible implementation, the control device 300 can be used for:

[0103] Based on the mapping relationship between the distorted coordinates of pixels in the degraded sample image and the corresponding pixel coordinates in the ground truth image, coordinate correction is performed on the degraded sample image to obtain a corrected degraded sample image aligned with the ground truth image; an image dataset is obtained based on the ground truth image and the corrected degraded sample image.

[0104] The mapping relationship between the distorted coordinates of pixels in the sample degraded image and the corresponding pixel coordinates in the ground truth image is the mapping relationship between the distorted coordinates of pixels in the distorted image and the corresponding pixel coordinates in the calibration image.

[0105] One way to perform coordinate correction on a degraded sample image is to remap the coordinates of the pixel positions in the degraded sample image based on the mapping relationship, so as to obtain a corrected degraded sample image that is aligned with the ground truth image.

[0106] Through the above methods, pixel-level alignment between ground truth images and degraded sample images in the entire image dataset can be achieved while ensuring the authenticity of the physical imaging link, thus providing an ideal data foundation for the supervised training of various computational imaging models.

[0107] The training method of the computational imaging model according to embodiments of this disclosure is described below. The computational imaging model of embodiments of this disclosure can be implemented using various deep learning architectures, including but not limited to multi-scale convolutional neural networks, Transformer-based architectures, and generative adversarial network architectures. This disclosure does not limit the specific network structure of the computational imaging model.

[0108] Figure 4 A schematic diagram illustrating the training and testing phases of a computational imaging model according to an embodiment of the present disclosure is shown. Figure 4 As shown, during the training phase, degraded images from the image dataset can be used as input to the computational imaging model. After processing, the model outputs the corresponding reconstructed image. The ground truth images corresponding to the degraded images can be used as the targets (labels) for supervised learning. A pre-defined loss function is used to calculate the loss function values ​​between the reconstructed and ground truth images. Based on these loss function values, the parameters of the computational imaging model are iteratively updated using the backpropagation algorithm until pre-defined training stopping conditions are met (such as loss function convergence, the number of training epochs reaching a pre-defined threshold, etc.), resulting in the trained computational imaging model. The essence of this training process is to enable the computational imaging model to learn a complex mapping function capable of reversing the degradation process of the entire physical imaging chain.

[0109] This disclosure does not limit the specific type of loss function. To achieve optimal image reconstruction quality, the loss function can be selected based on the training objective of the computational imaging model and the application scenario. For example, the loss function may include:

[0110] 1. Pixel-level loss functions: such as mean absolute error (MAE) loss or mean squared error (MSE) loss, are used to measure the difference in pixel intensity between the reconstructed image and the ground truth image, so as to constrain the consistency of the reconstruction results in brightness and color values.

[0111] 2. Structural loss functions: such as structural similarity (SSIM) loss or multi-scale structural similarity (MS-SSIM) loss, are used to constrain the reconstructed image to be consistent with the ground truth image in terms of brightness distribution, contrast relationship and structural information, thereby improving the structural fidelity of the reconstructed image.

[0112] 3. Chromaticity Loss Function: The chromaticity loss function can be used to solve color distortion problems that may occur during the imaging process. This chromaticity loss function can constrain the proportional relationship between different color channels in the reconstructed image by calculating the difference between the normalized chromaticity maps corresponding to the reconstructed image and the ground truth image. This allows the computational imaging model to recover a reasonable color distribution even when the brightness information is insufficient or saturated, thereby suppressing the generation of color-shift artifacts.

[0113] Traditional deep learning-based computational imaging models typically rely on training from scratch on massive datasets, which not only places high demands on computational resources and training time, but also often requires rebuilding the dataset and performing complete training when faced with new imaging hardware modules or specific application scenarios, making rapid adaptation difficult. To address these issues, embodiments of this disclosure introduce a "pre-training-fine-tuning" transfer learning mechanism in the aforementioned training phase to improve model training efficiency and scenario adaptability.

[0114] In one possible implementation, the control device 300 can be used for:

[0115] Without the optical degradation simulation device 500, the display device 100 and the image acquisition device 200 are controlled to acquire a first image dataset; with the optical degradation simulation device 500 in place, the display device 100 and the image acquisition device 200 are controlled to acquire a second image dataset.

[0116] Training the computational imaging model includes:

[0117] The initial computational imaging model is pre-trained using the first image dataset to obtain the pre-trained computational imaging model; the pre-trained computational imaging model is then fine-tuned using the second image dataset to obtain the fine-tuned computational imaging model.

[0118] The ground truth images in the first image dataset can be used to characterize ideal imaging results under general imaging scenarios. These ground truth images are, for example, natural scene images acquired under standard lighting conditions. By using the first image dataset for model pre-training, the computational imaging model learns basic image reconstruction capabilities.

[0119] The ground truth images in the second image dataset can be related to the target imaging scene targeted by the fine-tuning training. Target imaging scenes could be, for example, low-light conditions, haze scattering, or optical blurring. The imaging parameters adjusted by the optical degradation simulation device 500 are also related to the target imaging scene. For instance, in a low-light imaging scene, the ground truth image could be a night scene image, and the optical degradation simulation device 500 could be a neutral density filter. This allows the second image dataset to realistically reflect the physical imaging degradation characteristics of the target application scene.

[0120] Both pre-training and fine-tuning training methods can refer to the training process of the computational imaging model described above. The difference lies in that fine-tuning training is a further training process based on the pre-trained computational imaging model, and the image datasets used for pre-training and fine-tuning training are different. Since the computational imaging model already possesses basic image reconstruction capabilities in the pre-training stage, the fine-tuning stage usually requires only a few training iterations to achieve rapid convergence, thereby significantly reducing training time and computational costs, and specifically improving the model's reconstruction performance in specific imaging scenarios.

[0121] See Figure 4 During the testing phase, untrained degraded images can be input into a trained computational imaging model (e.g., the finely tuned computational imaging model described above). The degraded images can be test images acquired by the image acquisition device 200 using the system of this embodiment, or actual images captured by the image acquisition device 200 in real-world application scenarios. The computational imaging model can perform forward inference computation on the input degraded image and output a corresponding restored and enhanced high-quality image, i.e., the enhanced image described above. The reconstruction performance of the computational imaging model can be evaluated by comparing the differences in metrics between the enhanced image and the corresponding degraded image, or, if a ground truth image is available, by comparing the differences in metrics between the enhanced image and the ground truth image. These metrics can include peak signal-to-noise ratio (PSNR), sharpness, color deviation, etc., thereby achieving an objective evaluation of the model's restoration capability under real-world imaging degradation conditions.

[0122] Figure 5 A flowchart illustrating a method for generating an image dataset for computational imaging according to an embodiment of the present disclosure is shown. This method can be used in the aforementioned image dataset generation system for computational imaging, such as... Figure 5 As shown, the method includes:

[0123] Step S501: Display device 100 displays one or more frames of truth images;

[0124] In step S502, the image acquisition device 200 acquires images of the ground truth image to obtain sample degraded images that correspond one-to-one with each frame of the ground truth image.

[0125] In step S503, the control device 300, which is connected to the display device 100 and the image acquisition device 200 respectively, obtains an image dataset based on the ground truth image and the sample degraded image.

[0126] The degraded images in this dataset have lower imaging metrics than the corresponding ground truth images. These metrics include at least one of resolution, brightness, color, and signal-to-noise ratio. The image dataset is used to train a computational imaging model, enabling the trained model to process the degraded images acquired by the image acquisition device to obtain enhanced images. The enhanced images have higher imaging metrics than the degraded images.

[0127] In one possible implementation, step S503 includes:

[0128] Based on the mapping relationship between the distorted coordinates of pixels in the degraded sample image and the corresponding pixel coordinates in the ground truth image, coordinate correction is performed on the degraded sample image to obtain a corrected degraded sample image aligned with the ground truth image.

[0129] The image dataset is obtained based on the ground truth image and the corrected sample degraded image.

[0130] In one possible implementation, the method further includes:

[0131] In the absence of the optical degradation simulation device 500, the control device 300 controls the display device 100 and the image acquisition device 200 to acquire a first image dataset.

[0132] When the optical degradation simulation device 500 is placed, the control device 300 controls the display device 100 and the image acquisition device 200 to acquire a second image dataset.

[0133] Training the computational imaging model includes:

[0134] The initial computational imaging model was pre-trained using the first image dataset to obtain the pre-trained computational imaging model.

[0135] The pre-trained computational imaging model was fine-tuned using the second image dataset to obtain the fine-tuned computational imaging model.

[0136] Among them, the ground truth images in the second image dataset are related to the target imaging scene targeted by the fine-tuning training, and the imaging indicators adjusted by the optical degradation simulation device 500 are related to the target imaging scene.

[0137] According to embodiments of this disclosure, the ground truth image output by the display device can be used as a controllable, lossless source of truth. Corresponding degraded observation results (i.e., sample degraded images) acquired by the image acquisition device can be obtained through a real physical imaging link. This introduces the physical degradation effects generated by the real optical system and image sensor without relying on complex simulation modeling. This ensures that the ground truth image content is flexibly configurable while the generated sample degraded images accurately reflect the imaging characteristics of the target imaging device (i.e., the image acquisition device) under actual working conditions. Embodiments of this disclosure can generate a pairwise image dataset of "ground truth image - sample degraded image" with pixel-level correspondence under conditions of low equipment cost and high acquisition efficiency. This image dataset combines content diversity and physical realism, making it suitable for training and evaluating computational imaging models, thereby providing stable and scalable data support for the development of data-driven computational imaging algorithms.

[0138] In some embodiments, the functions or modules of the apparatus provided in this disclosure can be used to perform the methods described in the above method embodiments. The specific implementation can be referred to the description of the above method embodiments, and for the sake of brevity, it will not be repeated here.

[0139] This disclosure also provides an image dataset generation apparatus for computational imaging, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above method.

[0140] This disclosure also provides a non-volatile computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the above-described method.

[0141] This disclosure also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described method.

[0142] Computer-readable storage media can be tangible devices capable of holding and storing programs / instructions used by instruction execution devices. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination of the foregoing. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0143] The computer program (or computer-readable program instructions) described herein can be downloaded from a computer-readable storage medium to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage medium in the respective computing / processing device.

[0144] The computer program (or computer program instructions) used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing state information from the computer-readable program instructions to implement various aspects of this disclosure.

[0145] Various aspects of this disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0146] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that, when executed by the processor of the computer or other programmable data processing apparatus, they create means for implementing the functions / actions specified in one or more blocks of the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner; thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing aspects of the functions / actions specified in one or more blocks of the flowchart and / or block diagram.

[0147] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0148] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0149] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or technical improvements to the embodiments in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A system for generating image datasets for computational imaging, characterized in that, The system includes: Display devices are used to display one or more frames of truth images; An image acquisition device is used to acquire images of the ground truth image to obtain sample degraded images that correspond one-to-one with each frame of the ground truth image. The imaging index of the sample degraded image is lower than that of the corresponding ground truth image. The imaging index includes at least one of resolution, brightness, color, and signal-to-noise ratio. A control device connected to the display device and the image acquisition device respectively is used to obtain an image dataset based on the ground truth image and the sample degraded image. The image dataset is used to train a computational imaging model, so that the trained computational imaging model can process the degraded image to be processed acquired by the image acquisition device to obtain an enhanced image, wherein the imaging index of the enhanced image is higher than that of the degraded image to be processed.

2. The system according to claim 1, characterized in that, The system also includes a mechanical displacement device. The mechanical displacement device is configured to support the display device and / or the image acquisition device, and is used to adjust the relative position of the display device and the image acquisition device.

3. The system according to claim 2, characterized in that, The mechanical displacement device includes a robotic arm, which is used to control the movement of the image acquisition device according to preset motion parameters when the image acquisition device performs image acquisition. The preset motion parameters include one or more of motion trajectory, motion speed, acceleration and vibration frequency.

4. The system according to claim 1, characterized in that, The system also includes an optical degradation simulation device. The optical degradation simulation device is placed between the display device and the image acquisition device to reduce any one or more of the imaging indices of the sample degraded image relative to the true image.

5. The system according to claim 4, characterized in that, The optical degradation simulation device includes a neutral density filter used to attenuate the luminous flux passing through the imaging optical path, so that the brightness and / or signal-to-noise ratio of the sample degradation image is lower than that of the true image.

6. The system according to claim 4, characterized in that, The optical degradation simulation device includes a medium simulation chamber, which is used to introduce spectrally selective absorption and / or scattering effects into the imaging optical path so that the contrast and / or color shift of the sample degradation image is lower than that of the true image.

7. The system according to claim 1, characterized in that, The system also includes a light-shielding darkroom. The light-shielding dark box is placed outside the display device and the image acquisition device, and at least covers the imaging optical path area between the display device and the image acquisition device, in order to isolate external ambient light interference or simulate ambient lighting conditions when the image acquisition device is acquiring images.

8. The system according to claim 1, characterized in that, The control device is used for: Based on the mapping relationship between the distorted coordinates of pixels in the degraded sample image and the corresponding pixel coordinates in the ground truth image, coordinate correction is performed on the degraded sample image to obtain a corrected degraded sample image aligned with the ground truth image. An image dataset is obtained based on the ground truth image and the corrected sample degraded image.

9. The system according to any one of claims 1-8, characterized in that, The control device is used for: Without placing an optical degradation simulation device, the display device and the image acquisition device are controlled to acquire a first image dataset; With the optical degradation simulation device in place, the display device and the image acquisition device are controlled to acquire a second image dataset; The trained computational imaging model includes: The initial computational imaging model is pre-trained using the first image dataset to obtain the pre-trained computational imaging model. The pre-trained computational imaging model is fine-tuned using the second image dataset to obtain the fine-tuned computational imaging model. The ground truth images in the second image dataset are related to the target imaging scene for which the fine-tuning training is conducted, and the imaging metrics adjusted by the optical degradation simulation device are related to the target imaging scene.

10. A method for generating image datasets for computational imaging, characterized in that, The method includes: The display device displays one or more frames of truth images; The image acquisition device acquires images for the ground truth image to obtain sample degraded images that correspond one-to-one with each frame of the ground truth image. The imaging index of the sample degraded image is lower than that of the corresponding ground truth image. The imaging index includes at least one of resolution, brightness, color, and signal-to-noise ratio. A control device connected to the display device and the image acquisition device respectively obtains an image dataset based on the ground truth image and the sample degraded image. The image dataset is used to train a computational imaging model, so that the trained computational imaging model can process the degraded image to be processed acquired by the image acquisition device to obtain an enhanced image, wherein the imaging index of the enhanced image is higher than that of the degraded image to be processed.

Images