Strong backlight environment imaging sharpening system and method for industrial machine vision

Abstract

The invention discloses an industrial machine vision-oriented strong backlight environment imaging sharpening system and method. The method comprises the following steps: S10, performing optical coding imaging; s20, carrying out improved L-R iterative recovery and first-stage wavelet denoising; s30, carrying out secondary wavelet denoising and fine recovery; and S40, carrying out iterative convergence and output. According to the invention, a specific wavefront coding optical imaging module and an improved L-R image restoration algorithm module are integrated into a complete visual detection system for solving the problem of industrial strong backlight imaging. In the L-R iteration process, a staged denoising step based on wavelet transformation is introduced, and before each iteration, first-level high-threshold wavelet denoising is carried out on a residual error; and in a specific stage in the iteration process, carrying out secondary middle threshold wavelet denoising on the current estimation image.

Description

Technical Field

[0001] This invention belongs to the field of optical imaging technology, and specifically relates to a system and method for improving imaging clarity in strong backlight environments for industrial machine vision. Background Technology

[0002] In modern industrial automated production lines, machine vision systems play the role of "intelligent eyes," widely used in key areas such as product quality inspection, precision dimensional measurement, component positioning and identification, and robot guidance. The accuracy and reliability of these applications directly depend on the quality of the images acquired by the vision system. However, in many industrial settings, there exists a highly challenging imaging condition—strong backlighting environments. This environment typically occurs in the following scenarios: First, defect detection of transparent or semi-transparent objects such as glass bottles, transparent films, and LCD screens requires backlighting to highlight internal impurities, bubbles, or scratches; second, measuring the contour dimensions of metal parts and chip pins utilizes backlighting to produce a high-contrast silhouette effect, facilitating edge extraction; third, monitoring high-temperature processes such as welding and smelting, where the molten material itself becomes a very strong background light source; fourth, automated warehousing and logistics sorting in outdoor or strongly lit environments, where backlighting can cause target objects to appear dark and unclear.

[0003] In imaging under strong backlight conditions, the core challenge faced by industrial cameras lies in their insufficient dynamic range. The target object itself may have weak reflection or transmission, while the light source behind it is exceptionally strong. This results in a huge difference in the radiant energy received by each pixel unit on the sensor. As a result, two typical problems appear in the image: first, the target area appears dark and lacks detail due to underexposure, with blurred boundaries with the background; second, the background light source area produces "halos" or "glare" due to overexposure, and may even cause charge overflow, contaminating adjacent pixels and distorting the target outline. In addition, strong background light will also reflect multiple times within the optical system, introducing stray light and further reducing the overall contrast and signal-to-noise ratio of the image. These degradation effects severely interfere with subsequent image processing algorithms, such as edge detection, feature extraction, and pattern recognition, ultimately leading to decreased detection accuracy, increased false positive rate, and even complete failure of the vision system.

[0004] To address the challenges of high-backlight imaging, the industry has explored various technical approaches, which can be broadly categorized into hardware optimization and software algorithms.

[0005] Hardware optimization solution: High dynamic range imaging: This solution acquires multiple images of the same scene at different exposure times (e.g., a short exposure to capture bright details and a long exposure to capture dark details), and then combines them into a single high dynamic range image that retains both bright and dark information. This method extends the dynamic range to some extent, but its drawbacks include the need for multiple exposures, sensitivity to moving objects, potential ghosting, and a lengthy data processing flow, making it difficult to meet the demands of high-speed online detection.

[0006] Polarizing filter: A polarizing filter is installed in front of the lens to suppress glare from specific directions by rotating the polarization angle. This method is effective in eliminating reflections from metal surfaces, but its ability to solve the problem of an overly dark target and an overly bright background caused by differences in brightness limits is limited.

[0007] Active lighting design: By carefully designing lighting schemes, such as using coaxial lighting or dome lighting, direct light into the lens can be avoided as much as possible. However, this method is not applicable in scenarios where backlighting is necessary for contour detection.

[0008] Software Algorithm Scheme (Image Restoration): Inverse Filtering Algorithm: This is the most direct image restoration method. It assumes that the degradation process of the imaging system is linear and spatially invariant, and attempts to restore the original image by dividing the degraded image in the frequency domain by the Fourier transform of the point spread function (i.e., the optical transfer function). However, inverse filtering is extremely sensitive to noise because the division operation amplifies noise infinitely in frequency components where the point spread function value is close to zero, resulting in severe ringing effects in the restored image, making it very impractical.

[0009] Wiener filtering algorithm: As an improvement on inverse filtering, Wiener filtering introduces the statistical properties of noise and adds a regularization term to the inverse filtering algorithm, aiming to minimize the mean square error between the restored image and the original image. It outperforms inverse filtering in suppressing noise amplification. However, its performance heavily relies on prior knowledge estimation of the image and noise power spectra. In real industrial environments, noise characteristics are complex and variable, and accurate prior knowledge is difficult to obtain, leading to unstable restoration results.

[0010] Blind deconvolution algorithms: These algorithms estimate the point spread function (PSF) and the original image simultaneously when the PSF is unknown or partially unknown. They solve the problem iteratively by introducing prior constraints on the image (such as smoothness and sparsity). Although this reduces dependence on system parameters, blind deconvolution is inherently an ill-conditioned problem with non-unique solutions, prone to getting trapped in local optima, slow convergence, and sensitivity to initial values. In noisy environments, the recovery performance may deteriorate drastically.

[0011] Deep learning-based methods: In recent years, image restoration using models such as convolutional neural networks or Transformers has become a research hotspot. By training the model with a large number of paired degraded-sharp images, it can learn the complex mapping relationship for recovering details from poorly lit images. This approach has great potential, but its performance is highly dependent on the quality and quantity of training data. In industrial applications, acquiring a large number of realistic, paired degraded images with corresponding ideal sharp images in strong backlighting is costly. Furthermore, the model's generalization ability to scenes outside the training set may be insufficient.

[0012] In summary, existing technical solutions, whether hardware-level HDR or software-level traditional filtering and blind deconvolution, have varying degrees of limitations in terms of real-time performance, stability, adaptability, or implementation cost when dealing with complex, dynamic, and strong backlight environments in industrial settings.

[0013] The shortcomings of existing technologies include at least the following: Through in-depth analysis of existing technologies, we can summarize their main disadvantages in dealing with industrial strong backlight visual inspection: Cost and speed bottlenecks of hardware solutions: While HDR multi-exposure synthesis technology can extend the dynamic range, the process of multiple acquisitions and image fusion introduces time delays, which cannot meet the real-time detection requirements of high-speed pipelines. Special hardware or lighting designs increase the complexity and cost of the system.

[0014] Traditional algorithms are vulnerable to noise: inverse filtering is almost unusable in real-world noisy environments. The performance of Wiener filtering is limited by the accurate estimation of the statistical characteristics of noise, while noise sources in industrial environments are complex (such as sensor thermal noise, photon shot noise, quantization noise, etc.) and vary with ambient temperature and light intensity, making it difficult to describe with a fixed model.

[0015] Uncertainty and convergence issues in blind deconvolution: Due to its ill-conditioned nature, the blind deconvolution algorithm is unstable in its solution process, and the restored image heavily depends on the estimation of the initial point spread function and the setting of iteration parameters. Under strong backlighting and high-intensity noise interference, the algorithm is prone to diverging or converging to erroneous solutions that do not conform to physical meaning, resulting in severe artifacts or distortion in the restored image.

[0016] Dependency and Generalization Challenges of Deep Learning Methods: Data-driven deep learning methods require massive amounts of high-quality labeled data, and obtaining such a large and perfectly matched training set in specific industrial applications is extremely difficult. When a trained model faces a new scene, new workpiece, or new backlight intensity that is inconsistent with the distribution of the training data, its performance may degrade significantly, lacking sufficient robustness and generalization ability.

[0017] Insufficient ability to jointly suppress complex noise: Noise in strong backlight environments is not a single type; it is often a mixture of multiple noises. Most existing algorithms are designed for a specific type of noise and lack a comprehensive scheme that can synergistically suppress multiple noise mechanisms. This can lead to the amplification of one type of noise or the loss of valuable image details while suppressing another. Summary of the Invention

[0018] To address the shortcomings of the existing technologies, the present invention aims to provide a collaborative solution integrating optical modulation and advanced algorithm processing, fundamentally improving the imaging quality of industrial machine vision in strong backlight environments. The specific objectives are as follows: Constructing an optical imaging front-end resistant to strong backlight: Design an optical imaging system that incorporates wavefront coding technology. By introducing a specific phase mask in the optical path, the incident wavefront is encoded, so that the intermediate image formed on the sensor can naturally suppress the strong light diffusion effect, reduce the risk of saturation of individual pixels, and provide a more "friendly" input for subsequent algorithm processing.

[0019] Developing a robust and efficient image restoration algorithm: In the algorithm's backend, the classic Lucy-Richardson algorithm is innovatively improved to effectively handle complex noise in images with strong backlighting. The core lies in designing a dual-threshold denoising mechanism that integrates wavelet transform. During the iterative restoration process, noise is stripped away in stages and by type, thereby maximizing the suppression of noise amplification while restoring image details.

[0020] Achieve system-level collaborative optimization: Treat optical front-end modulation and algorithm back-end decoding as a complete system for joint design and optimization. Ensure that the coded blurring characteristics generated by the optical system are highly matched with the degradation model on which the algorithm relies, thereby achieving a "1+1>2" sharpening effect.

[0021] To meet the practical needs of industrial applications: The final system and method must have the characteristics of high real-time performance, strong robustness and good adaptability. It should be able to be stably applied to various complex industrial strong backlight detection scenarios without relying on a large amount of training data or significantly increasing hardware costs, and output target images with rich details and high contrast, so as to provide reliable visual basis for accurate automated decision-making.

[0022] This invention provides a system for enhancing image clarity in strong backlight environments for industrial machine vision, comprising a strong backlight-resistant optical imaging module, an image processing and restoration module, and a control and interface module connected in sequence. The anti-strong backlight optical imaging module acquires target images under strong backlight conditions; The image processing and restoration module receives the encoded intermediate image from the anti-strong backlight optical imaging module and executes the improved LR image restoration algorithm. The control and interface module transmits the recovered image to the host computer or robot controller via a network or bus interface.

[0023] Preferably, the strong backlight-resistant optical imaging module includes an industrial camera lens and a wavefront-coded phase mask.

[0024] Preferably, the image processing and restoration module is a GPU or FPGA.

[0025] To achieve the above objectives, the present invention also provides a method for improving image clarity in strong backlight environments for industrial machine vision, based on the above system, comprising the following steps: S10, optical coded imaging; S20, Improved LR Iterative Recovery and First-Level Wavelet Denoising; S30, Level 2 wavelet denoising and fine restoration; S40, Iterative convergence and output.

[0026] Preferably, S10 specifically involves, under strong backlight conditions, light from the target scene enters an anti-strong backlight optical imaging module integrated with a specific phase mask. This phase mask modulates the light wavefront, causing the point light source to no longer form an ideal point on the sensor, but instead form a known, diffused light spot with controlled energy distribution. The shape of this light spot is described by a point spread function; the entire scene forms a blurred intermediate image with known characteristics. The mathematical model is expressed as: , in, It is the desired original, clear image. It is a point spread function. Represents the convolution operation. This represents system noise introduced during the imaging process.

[0027] Preferably, S20 includes an intermediate image. and known point spread function As input, the iterative recovery process begins.

[0028] Preferably, in step S20, before each LR iteration, the residual from the previous iteration is subjected to first-level wavelet threshold denoising, and the denoised residual is denoised as... Then use The iterative formula for the calculations involved in the current iteration, after first-level denoising, is expressed as: , in, The recovered image is obtained after the k-th iteration. It is a point spread function. yes conjugate, It represents the number of iterations.

[0029] Preferably, step S30 includes processing the restored image obtained after the k-th iteration after completing a preset number of first-level denoising iterations. Two-stage wavelet denoising is performed using a multi-level decomposition of the wavelet basis and an intermediate threshold based on statistical properties. The wavelet coefficients are processed, and the image after two-stage denoising is denoted as... .

[0030] Preferably, step S40 denoises the image after secondary noise reduction. Using these as new initial values, continue executing S20 and S30 until the convergence condition is met, and finally output the image. This is the restored, clear target image.

[0031] The beneficial effects of the present invention include at least the following: 1. Suppressing strong light interference at the source: By using a wavefront-coded optical front end, this invention physically alters the imaging process, enabling the sensor to capture an unsaturated intermediate image with a more balanced information distribution. This provides a starting point for subsequent algorithmic processing that is far superior to the original strong backlight image, solving the fundamental problem of "input distortion" in traditional methods.

[0032] 2. Superior Noise Suppression Capability: This invention features a unique two-stage wavelet denoising mechanism embedded within the LR iterative framework, achieving precise noise suppression. The first-stage denoising rapidly removes most shot noise, while the second-stage denoising meticulously cleans up the mixture of residual noise and signal. This strategy effectively avoids the noise amplification problem of traditional LR algorithms and overcomes the contradiction of single-threshold denoising either leaving residual noise or damaging the signal. Experimental data shows that its PSNR and ISNR performance are improved by an average of over 7% and 6%, respectively, far exceeding the performance of comparable algorithms.

[0033] 3. Rich image detail and sharp edges: Due to effective noise control, the LR iterative algorithm can focus more on recovering the frequency components of the image, especially the high-frequency information representing details. The final recovered image not only has a low noise level, but also retains and enhances key features such as the texture and edges of the target to the greatest extent, which is very beneficial for subsequent machine vision analysis.

[0034] 4. Strong robustness and adaptability of the system: This method does not make strict assumptions about the type of noise and can adaptively handle mixed noise. The optical encoding process reduces the requirements for precise focusing and increases the system's depth of field. The entire scheme does not depend on specific large-scale datasets, and its performance is guaranteed by physical models and optimization algorithms. Therefore, it can exhibit stable performance when facing different industrial workpieces and varying backlight intensities.

[0035] 5. Balances real-time performance and efficiency: Compared to multi-exposure HDR and training complex deep learning models, the algorithm of this invention has a fixed flow and controllable computational load. Combined with modern parallel processing hardware (such as GPUs), it can meet the real-time requirements of most industrial online inspections, achieving a balance between high performance and high efficiency. Attached Figure Description

[0036] To make the objectives, technical solutions, and beneficial effects of this invention clearer, the following figures are provided for illustration: Figure 1 This is a schematic diagram of the imaging enhancement system for strong backlight environments for industrial machine vision, according to an embodiment of the present invention. Figure 2 This is a flowchart illustrating the image sharpening method for strong backlight environments for industrial machine vision, according to an embodiment of the present invention. Figure 3 This is a schematic diagram of the point spread function (PSF) and optical coding effect of a strong backlight environment imaging sharpening system for industrial machine vision according to an embodiment of the present invention. Figure 4 This is a comparison of the imaging and restoration effects of the image sharpening method for strong backlight environments in industrial machine vision, according to an embodiment of the present invention. Detailed Implementation

[0037] The preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0038] See Figure 1 A system for enhancing image clarity in strong backlight environments for industrial machine vision includes a strong backlight-resistant optical imaging module 10, an image processing and restoration module 20, and a control and interface module 30 connected in sequence. The strong backlight-resistant optical imaging module 10 acquires target images under strong backlight conditions. The image processing and restoration module 20 receives the encoded intermediate image from the strong backlight-resistant optical imaging module 10 and executes an improved LR image restoration algorithm. The control and interface module 30 transmits the restored image to a host computer or robot controller via a network or bus interface.

[0039] The strong backlight-resistant optical imaging module 10 includes an industrial camera lens and a wavefront-coded phase mask. This module is the physical front end of the system and is responsible for acquiring target images under strong backlight conditions. It consists of an industrial camera lens and an aspherical phase mask integrated inside the lens or at a key location in the optical path. This mask is the core of the wavefront coding technology. Its surface is precisely designed and lacks the rotational symmetry of traditional lenses, enabling it to specifically modulate (encode) the wavefront of the passing light waves. When light from a strong backlight scene passes through this module, the image formed on the camera sensor is not a traditional sharp image or a simple blurry image, but rather an "encoded intermediate image" that is insensitive to depth of field and defocus, and where the strong light energy is redistributed. This image appears blurry, but it faithfully records the modulated light field information, effectively suppressing the energy peaks in the strong background light areas, avoiding sensor saturation, and preserving the key information of the target object.

[0040] The image processing and restoration module 20 is a GPU or FPGA. This module is the "brain" of the system and is typically handled by an industrial computer or embedded high-performance processor (such as a GPU or FPGA). It receives the encoded intermediate image from the camera and runs the improved LR image restoration algorithm at the core of this invention. This module is responsible for controlling the entire restoration process, including calling the point spread function, executing iterative operations, implementing wavelet transform and threshold denoising, and finally outputting a cleared target image.

[0041] The control and interface module 30 is responsible for coordination within and outside the system. It generates synchronization signals to coordinate lighting (if a controllable light source is used), image acquisition and processing timing; at the same time, it transmits the recovered clear image to the host computer or robot controller via a network or bus interface (such as GigE, USB3.0, IO-link, etc.) to complete subsequent detection, measurement or guidance tasks.

[0042] See Figure 2 This is a flowchart of the improved LR image restoration algorithm of the present invention, which describes the algorithm process in detail in the form of a block diagram. Starting from "Input encoded intermediate image G and PSF (h)", the process enters "Initialize estimated image M(0)", and then enters the main loop: "First-level wavelet denoising (high threshold)" → "LR iterative update" → "Determine if the second-level denoising condition has been met?" If yes, perform "Second-level wavelet denoising (medium threshold)", otherwise continue the loop; "Determine if convergence has occurred?" If yes, "Output clear image M_final", otherwise continue the loop. Arrows indicate the direction of data and flow.

[0043] The specific method of this invention includes the following steps: S10, Optical Coding Imaging; specifically, in a strong backlight environment, light from the target scene enters an anti-strong backlight optical imaging module integrated with a specific phase mask. This phase mask modulates the light wavefront, causing the point light source to no longer form an ideal point on the sensor, but rather a known, diffused, but energy-controlled light spot. The shape of this light spot is described by a point spread function; the entire scene forms a blurred intermediate image with known characteristics. The mathematical model is expressed as: , in, It is the desired original, clear image. It is a point spread function. Represents the convolution operation. This represents system noise introduced during the imaging process.

[0044] S20, Improved LR Iterative Recovery with First-Level Wavelet Denoising; Improved LR Algorithm with this intermediate image and known point spread function As input, the iterative recovery process begins. Traditional LR algorithms, based on Bayes' theorem and maximum likelihood estimation, approximate the original image iteratively. Its basic iterative formula is: ,in yes conjugate, This refers to the number of iterations. However, in a noisy environment, the residual generated in each iteration... It contains a lot of noise Using it directly in the next iteration will cause the noise to be amplified rapidly.

[0045] The first innovation of this invention lies in performing a first-level wavelet thresholding denoising on the residuals of the previous iteration before each LR iteration. We analyze the residuals... The wavelet coefficients of noise are typically small in amplitude and widely distributed, while the wavelet coefficients of image signals (especially details such as edges) have larger amplitudes and are relatively concentrated. We set a high hard threshold, directly setting wavelet coefficients with amplitudes below this threshold to zero. This effectively removes most of the transform coefficients of random noise (type I noise). The denoised residual is denoted as... Then use The calculations involved in this iteration. The iterative formula after the first stage of denoising can be expressed as: This step significantly suppresses noise accumulation during the iteration process and accelerates convergence.

[0046] S30, Second-level wavelet denoising and fine restoration; After the first-level denoising, most of the obvious noise has been removed, but a small number of noise coefficients are still mixed with weak image signal coefficients (type II noise), which is difficult to completely separate with a high threshold. If the threshold is too low, it will damage the signal; if the threshold is too high, residual noise will remain.

[0047] Therefore, this invention introduces a second innovation: after completing a preset number of (e.g., 5) first-level denoising iterations, the currently estimated image is... The original method performed secondary wavelet denoising. This time, we chose a more refined wavelet basis (such as "db4") for deeper decomposition (e.g., 5 levels). We also employed an intermediate threshold based on statistical characteristics (e.g., adaptively calculated based on the median of the absolute values ​​of the coefficients in each sub-band) to denoise the data. The wavelet coefficients are processed. This intermediate threshold can more finely separate residual noise mixed in the signal. The image after secondary denoising is denoted as... .

[0048] S40, iterative convergence and output, outputting the image after secondary denoising. Using these as new initial values, continue executing steps two and three until the convergence condition is met (e.g., the difference between images from two adjacent iterations is less than a certain threshold, or the maximum number of iterations is reached). The final output image... This results in a clear target image after restoration. This strategy, which combines iterative denoising with staged fine denoising, achieves a remarkable balance between restoring details and suppressing noise by peeling away layers of complex noise.

[0049] See Figure 3 The diagram illustrates the point spread function (PSF) and optical coding effect. (a) The sub-image shows the PSF of a conventional optical system, which is a concentrated bright spot. (b) The sub-image shows the PSF of the wavefront coding system of this invention, which is a diffused, uniformly distributed special pattern. (c) The sub-image compares the effects of conventional imaging and coded imaging under strong light; the former has a large halo, while the latter's halo is effectively suppressed.

[0050] See Figure 4 The comparison chart of imaging and restoration results includes multiple sub-images, showing: (a) the target image under ideal conditions (reference); (b) the degraded image captured by a traditional camera under strong backlight (dark target, overexposed background, halo); (c) the coded intermediate image captured by the optical imaging module of this invention (overall blurry but no overexposure); (d) the restoration result of traditional inverse filtering (amplified noise, severe ringing); (e) the restoration result of Wiener filtering (still blurry and noisy); and (f) the final restoration result of the method of this invention (clear target, rich details, clean background).

[0051] The scope of protection of this invention is defined by a series of key technical points, which are the core of achieving excellent results: The synergistic system architecture of optics and algorithms: The protection point lies in integrating a specific wavefront-coded optical imaging module (characterized by a built-in phase mask for suppressing strong light diffusion) with an improved LR image restoration algorithm module into a complete visual inspection system to solve the problem of industrial strong backlight imaging.

[0052] The improved LR restoration algorithm flow: This method incorporates a phased denoising step based on wavelet transform during the LR iteration process. Specifically, it includes a complete flow of performing a first-level high-threshold wavelet denoising on the residual before each iteration, and a second-level medium-threshold wavelet denoising on the currently estimated image at specific stages during the iteration process.

[0053] Dual-threshold denoising strategy: This refers to the specific technical means of using two thresholds with different characteristics (a first-level high threshold is mainly used to remove the first type of noise figure with a small amplitude; a second-level medium threshold is used to finely separate the second type of noise figure mixed with the signal) to synergistically suppress noise.

[0054] Parameter optimization for industrial strong backlight scenarios: The method includes optimized parameter settings to adapt to the characteristics of industrial strong backlight images, such as the selection of wavelet basis functions (e.g., 'db4'), the number of wavelet decomposition layers (e.g., 5 layers), and the criteria for setting the iterative convergence conditions.

[0055] Acquisition and application of the point spread function: In the system described above, the wavefront coding system is calibrated to obtain its dedicated point spread function h, and this known h is used as a core parameter in a specific implementation method to improve the iterative process of the LR algorithm.

[0056] Specific applications in industrial inspection: This involves applying the above system and methods to industrial machine vision, specifically for defect detection, dimensional measurement, or positioning identification of transparent workpieces, metal contours, and high-temperature objects under strong backlight conditions.

[0057] In specific embodiments, the following alternative technical solutions may also be adopted: While adhering to the core principles of this invention, those skilled in the art may conceive of some alternative or equivalent modifications, which should also be considered as extensions of the scope of protection of this invention: Alternatives to phase mask design: The core is to generate a PSF that can suppress strong light diffusion and form a stable coded image. In addition to the aspherical form mentioned in the text, phase masks can also adopt other surface shapes such as cubic surfaces and freeform surfaces, or use liquid crystal spatial light modulators to dynamically generate different coded wavefronts to adapt to various scenarios.

[0058] Alternatives to the denoising transform domain: While wavelet transform is an excellent tool for analyzing non-stationary signals, it is not the only option. In the secondary denoising stage, multi-scale geometric analysis tools such as curvelet transform and contourlet transform, which are better able to capture the geometric features of image edges, can be considered to replace or partially replace wavelet transform in order to achieve better detail preservation.

[0059] Alternative threshold selection strategies: The high and medium thresholds mentioned in the paper can be fixed values ​​or adaptive thresholds based on image statistical characteristics. For example, an adaptive threshold calculation method can be used to dynamically determine the threshold based on the coefficient distribution of each wavelet sub-band, potentially achieving better denoising performance.

[0060] Fusion and enhancement of iterative algorithms: The LR algorithm is a classic choice. It can also be explored to combine it with regularization methods such as total variation minimization, simultaneously constraining image smoothness during iteration to further suppress ringing effects and enhance edges. Alternatively, faster optimization algorithms such as the conjugate gradient method can be used to accelerate iterative convergence.

[0061] Alternative processing platforms: To meet the requirements of extreme real-time performance, the entire improved LR image restoration algorithm, especially wavelet transform and convolution operations, can be executed on an FPGA or application-specific integrated circuit to achieve hardware acceleration and meet the microsecond-level processing speed requirements.

[0062] This invention effectively solves the imaging challenges faced by industrial machine vision in strong backlight environments by constructing a collaborative innovation system of "optical encoding-algorithm decoding". The system and method combine the robustness of front-end hardware with the intelligence of back-end algorithms, achieving effective suppression of complex noise and high-quality recovery of image details without excessively increasing costs, providing strong technical support for improving the reliability and accuracy of industrial automated inspection.

[0063] Finally, it should be noted that the above preferred embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail through the above preferred embodiments, those skilled in the art should understand that various changes can be made to it in form and detail without departing from the scope defined by the claims of the present invention.

Claims

1. A system for enhancing imaging clarity in strong backlight environments for industrial machine vision, characterized in that, It includes a strong backlight-resistant optical imaging module, an image processing and restoration module, and a control and interface module connected in sequence, wherein, The anti-strong backlight optical imaging module acquires target images under strong backlight conditions; The image processing and restoration module receives the encoded intermediate image from the anti-strong backlight optical imaging module and executes the improved LR image restoration algorithm. The control and interface module transmits the recovered image to the host computer or robot controller via a network or bus interface.

2. The image sharpening system for strong backlight environments for industrial machine vision according to claim 1, characterized in that, The anti-strong backlight optical imaging module includes an industrial camera lens and a wavefront-coded phase mask.

3. The image sharpening system for strong backlight environments for industrial machine vision according to claim 1, characterized in that, The image processing and restoration module is a GPU or FPGA.

4. A method for improving image clarity in strong backlight environments for industrial machine vision, characterized in that, The system based on any one of claims 1-3 includes the following steps: S10, optical coded imaging; S20, Improved LR Iterative Recovery and First-Level Wavelet Denoising; S30, Level 2 wavelet denoising and fine restoration; S40, Iterative convergence and output.

5. The method for improving image clarity in strong backlight environments for industrial machine vision according to claim 4, characterized in that, Specifically, in a strong backlight environment, light from the target scene enters an anti-strong backlight optical imaging module integrated with a specific phase mask. This phase mask modulates the light wavefront, causing the point light source to no longer form an ideal point on the sensor, but instead form a known, diffused light spot with controlled energy distribution. The shape of this light spot is described by a point spread function; the entire scene forms a blurred intermediate image with known characteristics. The mathematical model is expressed as: ， in, It is the desired original, clear image. It is a point spread function. Represents the convolution operation. This represents system noise introduced during the imaging process.

6. The method for improving image clarity in strong backlight environments for industrial machine vision according to claim 5, characterized in that, S20 includes an intermediate image. and known point spread function As input, the iterative recovery process begins.

7. The method for improving image clarity in strong backlight environments for industrial machine vision according to claim 6, characterized in that, In step S20, before each LR iteration, the residual from the previous iteration is subjected to first-level wavelet threshold denoising, and the denoised residual is denoised as... Then use The iterative formula for the calculations involved in the current iteration, after first-level denoising, is expressed as: ， in, The recovered image is obtained after the k-th iteration. It is a point spread function. yes conjugate, It represents the number of iterations.

8. The method for improving image clarity in strong backlight environments for industrial machine vision according to claim 7, characterized in that, S30 includes, after completing a preset number of first-level denoising iterations, processing the restored image obtained after the k-th iteration. Two-stage wavelet denoising is performed using a multi-level decomposition of the wavelet basis and an intermediate threshold based on statistical properties. The wavelet coefficients are processed, and the image after two-stage denoising is denoted as... .

9. The method for improving image clarity in strong backlight environments for industrial machine vision according to claim 8, characterized in that, S40 processes the image after secondary noise reduction. Using these as new initial values, continue executing S20 and S30 until the convergence condition is met, and finally output the image. This is the restored, clear target image.

Images