High-precision color identification method for complex light source environment

Abstract

The invention discloses a high-precision color identification method for a complex light source environment, and relates to the technical field of color identification. The method specifically comprises the following steps: 1) constructing a hardware system; 2) carrying out image preprocessing; 3) extracting image features by using an improved YOLO algorithm, performing multi-scale feature fusion by using a feature pyramid network, and correcting color distortion caused by a light source, equipment or a material in combination with a color restoration algorithm to ensure real restoration of the color; 4) designing a color evaluation algorithm, outputting color parameters such as a Lab value and a delta E value, accurately evaluating the color, and performing color difference comparison; 5) establishing a standard database, supporting the learning and optimization of historical data, and improving the identification accuracy of a new sample; 6) an automatic operation process; according to the method, advanced hardware configuration, deep learning, color restoration and automatic control technologies are combined, high-precision color identification in a complex light source environment is facilitated, and the method has a wide application prospect.

Description

Technical Field

[0001] The present invention relates to the technical field of color recognition, and in particular to a high-precision color recognition method for complex light source environments. Background Art

[0002] In traditional color recognition methods, changes in lighting conditions often lead to a decrease in color recognition accuracy, especially in complex lighting environments, where color distortion is particularly prominent. Different light sources have different spectral characteristics, which can significantly affect the color reflected from an object's surface, leading to inaccurate color recognition. Furthermore, most existing color recognition technologies are based on simple image processing algorithms, which often cannot effectively adapt to multiple light sources or uneven lighting environments, resulting in large errors in recognition results.

[0003] In fields such as industrial inspection, surface defect detection for smart appliances, and color design, high-precision color recognition is required. However, existing technologies lack adaptability and robustness to meet the actual needs of these application scenarios. Therefore, there is an urgent need for a high-precision color recognition method that can quickly and accurately identify colors in complex lighting environments while ensuring true reproduction. Summary of the Invention

[0004] The purpose of the present invention is to address the above-mentioned shortcomings and propose a high-precision color recognition method for complex light source environments. By adopting advanced hardware systems, image processing technology and deep learning algorithms, this method can quickly and accurately identify the color of objects under different lighting conditions and ensure the true restoration of colors.

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

[0006] A high-precision color recognition method for complex light source environments specifically includes the following steps:

[0007] S1. Hardware system construction;

[0008] S2, preprocessing the collected images, including denoising, contrast enhancement and color balance;

[0009] The original image is I, and the denoised image is I′, then:

[0010] I′=D(I)

[0011] Where D represents the contrast enhancement of the denoising algorithm. Assuming that the original contrast of the image is C and the enhanced image is I″, then:

[0012] I″=α(I′-μ)+β, where α is the contrast enhancement coefficient, μ is the mean value of the image, and β is the brightness offset;

[0013] Color balance: Ensures the image presents true colors by adjusting the color distribution of RGB channels;

[0014] S3, image feature extraction and color distortion correction;

[0015] During the feature extraction process, the YOLO algorithm is used to detect the target in the image, quickly extract the feature information of the target object, and fuse the multi-scale feature information through the feature pyramid network to improve the color recognition accuracy in complex backgrounds;

[0016] In order to correct the color distortion caused by light source, equipment or material, we combine the color restoration algorithm to analyze the impact of different light sources on the image color and use spectral correction to correct the image color.

[0017] Feature extraction: Use the improved YOLO algorithm to extract features from the image. Assuming that the input image is I and the extracted feature map is F, then:

[0018] F=YOLO(I)

[0019] Feature fusion: Multi-scale features are fused through the feature pyramid network FPN to generate the fused feature map F′:

[0020] F′=FPN(F)

[0021] Color distortion correction: Correct the color distortion caused by the light source through the color restoration algorithm. Assume that the color of the input image is C in , the corrected color is C out ,but:

[0022] C out =CR(C in )

[0023] Among them, CR represents the color restoration algorithm;

[0024] S4, color evaluation and color difference calculation;

[0025] Design a color evaluation algorithm and use Lab value and ΔE value to accurately evaluate the color.

[0026] Lab value calculation: Assuming the RGB value in the image is (R, G, B), its corresponding Lab value is (L, a, b), which can be calculated using the following conversion formula:

[0027]

[0028] b=200[f(Y / Y n )-f(Z / Z n )]

[0029] Among them, X n ,Y n ,Z n is the XYZ value of the reference white point,

[0030]

[0031] ΔE value calculation: ΔE value is a standard used to measure the difference between two colors. The calculation formula is:

[0032]

[0033] By calculating the ΔE value, the color error is accurately evaluated and the color accuracy is judged according to the set threshold;

[0034] S5. Standard database construction and optimization;

[0035] S6, automated operation process;

[0036] S7. Model optimization and real-time data update.

[0037] Preferably, the hardware system in S1 is constructed using an area array camera, a liquid lens, and a laser camera, and provides uniform and stable illumination through a parallel light source;

[0038] Area array cameras are used to acquire high-resolution image data. Liquid lenses can adjust the focal length according to the size and position of different objects to ensure image clarity. Laser cameras provide additional depth information to enhance the ability to capture image details. Parallel light sources eliminate the impact of light source unevenness on the image through uniform lighting conditions, thereby ensuring the authenticity of color reproduction.

[0039] Preferably, in the construction and optimization of the standard database in S5, a standard database is constructed to store standard color data under different light sources. The data in the database is calibrated and verified to ensure its accuracy. By learning and optimizing historical data, the color recognition accuracy of new samples is continuously improved. The construction process of the standard database includes:

[0040] 4) Collect standard color data under different light sources;

[0041] 5) Calibrate and verify the data to ensure its accuracy;

[0042] 6) Use machine learning algorithms to optimize data and improve the quality and reliability of the database.

[0043] The automated operation process in S6 realizes the full process automation from sample input to final result display. Users can complete all tasks through simple interface settings; it automatically completes image acquisition, processing, feature extraction, color correction, color evaluation, and finally outputs a detailed identification report. The specific process is as follows:

[0044] 5) The user enters sample information;

[0045] 6) The system automatically collects images and performs preprocessing;

[0046] 7) The system performs feature extraction, color correction and color evaluation on the image;

[0047] 8) Output the results and generate a color difference report.

[0048] Preferably, the specific optimization process of model optimization and real-time data update in S7 is as follows:

[0049] Model training: Based on the enhanced training data set, the deep learning model is used for training. The improved YOLO model is selected to obtain the feature map F of the sample image. The feature map is processed using the improved feature pyramid network and then feature fusion F′ is performed. The detection head is used to analyze the feature map and predict the target category and position coordinates of each detection frame. The detection head also predicts the confidence P of each detection frame, which indicates the probability that the detection frame contains the target. A confidence threshold T is set. Only detection frames with a confidence higher than the threshold will be judged as having defects.

[0050] If P>T, it is determined that there is a defect

[0051] Cross-validation: During the training process, the model is trained and validated multiple times using the cross-validation method to ensure the generalization ability of the model;

[0052] Performance evaluation: Use the test data set to test the trained model and evaluate the performance of the model. By calculating the accuracy A, recall R, and F1 value F1, the detection effect of the model is comprehensively evaluated;

[0053]

[0054] Among them, TP represents true positive examples, FP represents false positive examples, and FN represents false negative examples.

[0055] The present invention has the following effects:

[0056] By utilizing advanced hardware systems, image processing techniques, and deep learning algorithms, this method can quickly and accurately identify the color of objects under varying lighting conditions, ensuring true color reproduction. This method not only improves color recognition accuracy but also enhances adaptability in complex environments. It has broad application prospects, particularly in industrial inspection, surface defect detection for smart appliances, and color design. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 This is the system architecture diagram of this application;

[0058] Figure 2 Construct and optimize flow charts for standard databases;

[0059] Figure 3 It is a flowchart for automated operations;

[0060] Figure 4 Flowchart for model optimization and real-time data update. DETAILED DESCRIPTION

[0061] The specific implementation of the present invention will be further described below with reference to the accompanying drawings and specific embodiments:

[0062] Combine Figure 1 A high-precision color recognition method for complex light source environments includes the following steps:

[0063] S1. Hardware system construction, which uses an area array camera, liquid lens and laser camera, and provides uniform and stable lighting through a parallel light source.

[0064] Area array cameras are used to acquire high-resolution image data. Liquid lenses can adjust the focal length according to the size and position of different objects to ensure image clarity. Laser cameras provide additional depth information to enhance the ability to capture image details. Parallel light sources eliminate the impact of light source unevenness on the image through uniform lighting conditions, thereby ensuring the authenticity of color reproduction.

[0065] S2, pre-process the collected images, such as Figure 2 As shown, it includes denoising, contrast enhancement, and color balancing;

[0066] The original image is I, and the denoised image is I′, then:

[0067] I′=D(I)

[0068] Where D represents the contrast enhancement of the denoising algorithm. Assuming that the original contrast of the image is C and the enhanced image is I″, then:

[0069] I″=α(I′-μ)+β

[0070] Among them, α is the contrast enhancement coefficient, μ is the mean of the image, and β is the brightness offset;

[0071] Color balance: Ensures the image presents true colors by adjusting the color distribution of RGB channels;

[0072] S3, image feature extraction and color distortion correction;

[0073] During the feature extraction process, the YOLO algorithm is used to detect targets in the image, quickly extract the feature information of the target object, and fuse the multi-scale feature information through the feature pyramid network to improve the color recognition accuracy in complex backgrounds.

[0074] In order to correct the color distortion caused by light source, equipment or material, we combine the color restoration algorithm to analyze the impact of different light sources on the image color and use spectral correction to correct the image color.

[0075] Feature extraction: Use the improved YOLO algorithm to extract features from the image. Assuming that the input image is I and the extracted feature map is F, then:

[0076] F=YOLO(I)

[0077] Feature fusion: Multi-scale features are fused through the feature pyramid network FPN to generate the fused feature map F′:

[0078] F′=FPN(F)

[0079] Color distortion correction: Correct the color distortion caused by the light source through the color restoration algorithm. Assume that the color of the input image is C in , the corrected color is C out ,but:

[0080] C out =CR(C in )

[0081] Among them, CR represents the color restoration algorithm;

[0082] S4, color evaluation and color difference calculation;

[0083] Design a color evaluation algorithm and use Lab value and ΔE value to accurately evaluate the color.

[0084] Lab value calculation: Assuming that the RGB value in the image is (R, G, B), its corresponding Lab value is (L, a, b), which can be calculated using the following conversion formula:

[0085]

[0086] b=200[f(Y / Yn )-f(Z / Z n )]

[0087] Among them, X n , Y n , Z n is the XYZ value of the reference white point,

[0088]

[0089] ΔE value calculation: ΔE value is a standard used to measure the difference between two colors. The calculation formula is:

[0090]

[0091] By calculating the ΔE value, the color error is accurately evaluated and the color accuracy is judged according to the set threshold;

[0092] S5, Standard database construction and optimization Figure 2 As shown in the figure, by building a standard database, standard color data under different light sources is stored. The data in the database is calibrated and verified to ensure its accuracy. By learning and optimizing historical data, the color recognition accuracy of new samples is continuously improved. The construction process of the standard database includes:

[0093] 7) Collect standard color data under different light sources;

[0094] 8) Calibrate and verify the data to ensure its accuracy;

[0095] 9) Use machine learning algorithms to optimize data and improve the quality and reliability of the database.

[0096] S6, automated operation process, such as Figure 3 As shown, the entire process from sample entry to final result display is automated. Users can complete all tasks through simple interface settings; image acquisition, processing, feature extraction, color correction, color evaluation are automatically completed, and a detailed identification report is finally output. The specific process is as follows:

[0097] 9) The user enters sample information;

[0098] 10) The system automatically collects images and performs preprocessing;

[0099] 11) The system performs feature extraction, color correction and color evaluation on the image;

[0100] 12) Output the results and generate a color difference report.

[0101] S7, model optimization and real-time data update, such as Figure 4 The specific optimization process is shown as follows:

[0102] Model training: Based on the enhanced training data set, the deep learning model is used for training. The improved YOLO model is selected to obtain the feature map F of the sample image. The feature map is processed using the improved feature pyramid network and then feature fusion F′ is performed. The detection head is used to analyze the feature map and predict the target category and position coordinates of each detection frame. The detection head also predicts the confidence P of each detection frame, which indicates the probability that the detection frame contains the target. A confidence threshold T is set. Only detection frames with a confidence higher than the threshold will be judged as having defects.

[0103] If P>T, it is determined that there is a defect

[0104] Cross-validation: During the training process, the model is trained and validated multiple times using the cross-validation method to ensure the generalization ability of the model;

[0105] Performance evaluation: Use the test data set to test the trained model and evaluate the performance of the model. By calculating the accuracy A, recall R, and F1 value F1, the detection effect of the model is comprehensively evaluated;

[0106]

[0107]

[0108] Among them, TP represents true positive examples, FP represents false positive examples, and FN represents false negative examples.

[0109] To meet the color designers' demand for faithful color reproduction, an integrated hardware system was specially designed. This system not only includes a high-resolution area array camera, liquid lens, and laser camera, but also features an intelligent lighting control system. This control system automatically adjusts color temperature and brightness based on the user's selected light source type (such as daylight, incandescent lamp, etc.), simulating various real-world usage scenarios to ensure consistent and authentic image acquisition. Furthermore, the use of light-absorbing materials further reduces interference from external light sources and improves image quality. The system also supports multiple connection methods, facilitating seamless integration with commonly used tools and platforms for designers.

[0110] Step 2: Image preprocessing

[0111] In color design, the key to image preprocessing is to maintain the authenticity of the original colors and the clarity of details. Therefore, a deep learning-based denoising algorithm and adaptive contrast enhancement technology are used. For each input image I, a deep learning noise removal network D is first applied for denoising:

[0112] I′=D(I)

[0113] Then, the contrast and brightness are dynamically adjusted based on the image content to ensure the most realistic color information in the image. Furthermore, a color balance module B is introduced to automatically correct the RGB channels based on different material properties, making the colors of the final output image I" more realistic.

[0114] Step 3: Image feature extraction and color distortion correction

[0115] In view of the characteristics of color design, we improved the YOLO algorithm to better capture subtle color changes and combined it with the Feature Pyramid Network (FPN) to achieve multi-scale feature fusion. Assuming that the input image is I and the feature map after feature extraction is F, we have:

[0116] F=YOLO(I)

[0117] Next, FPN is used to fuse the multi-scale features to generate the fused feature map F′:

[0118] F′=FPN(F)

[0119] On this basis, the color restoration algorithm CR using spectral correction technology corrects the color distortion caused by the light source, ensuring that the design works can present consistent color effects under different light sources:

[0120] C out =CR(C in )

[0121] Step 4: Color evaluation and color difference calculation

[0122] To help designers accurately evaluate color differences, a color evaluation system based on the Lab color space has been designed. By calculating the Lab value (L, a, b) and the ΔE value, the visual difference between two colors can be quantified. The specific formula is as follows:

[0123]

[0124] b=200[f(Y / Y n )-f(Z / Z n )]

[0125] Among them, X n , Y n , Z n is the XYZ value of the reference white point,

[0126]

[0127] The calculation formula for color difference ΔE is:

[0128]

[0129] Step 5: Standard database construction and optimization

[0130] We've established a standard database containing standard color data for various lighting environments. This data covers not only common light sources but also includes color samples commonly used by designers and their performance on different materials. By learning from and optimizing historical data, we continuously improve the color recognition accuracy of new samples. The database supports dynamic updates to reflect the latest color trends and technological developments.

[0131] Step 6: Automated Operation Process

[0132] To improve work efficiency, the system automates the entire process, from sample entry to final result display. Designers simply set parameters through a user-friendly interface, and the system automatically completes tasks such as image acquisition, preprocessing, feature extraction, color correction, and color evaluation, generating detailed identification reports. Furthermore, the system provides cloud storage, allowing designers to easily access and manage project files at any time.

[0133] Step 7: Model optimization and real-time data update

[0134] By continuously collecting new image data and adding it to the training set, the model is regularly retrained and its parameters adjusted to ensure it can adapt to changing design needs. To address the specific requirements of color design, the model's generalization capabilities for small sample sizes have been optimized, enabling designers to obtain reliable predictions in a short period of time. Furthermore, the system supports online updates, ensuring users always have access to the latest versions of algorithms and technologies.

[0135] Of course, the above description is not a limitation of the present invention, and the present invention is not limited to the above examples. Changes, modifications, additions or substitutions made by technicians in this technical field within the essential scope of the present invention should also fall within the scope of protection of the present invention.

Claims

1. A high-precision color recognition method for complex light source environments, characterized in that: The specific steps include: S1. Hardware system construction; S2, preprocessing the collected images, including denoising, contrast enhancement and color balance; The original image is I, and the denoised image is I′, then: I′=D(I) Where D represents the contrast enhancement of the denoising algorithm. Assuming that the original contrast of the image is C and the enhanced image is I″, then: I"=a(I'-μ)+β Among them, α is the contrast enhancement coefficient, μ is the mean of the image, and β is the brightness offset; Color balance: Ensures the image presents true colors by adjusting the color distribution of RGB channels; S3, image feature extraction and color distortion correction; During the feature extraction process, the YOLO algorithm is used to detect the target in the image, quickly extract the feature information of the target object, and fuse the multi-scale feature information through the feature pyramid network to improve the color recognition accuracy in complex backgrounds; In order to correct the color distortion caused by light source, equipment or material, we combine the color restoration algorithm to analyze the impact of different light sources on the image color and use spectral correction to correct the image color. Feature extraction: Use the improved YOLO algorithm to extract features from the image. Assuming that the input image is I and the extracted feature map is F, then: F=YOLO(I) Feature fusion: Multi-scale features are fused through the feature pyramid network FPN to generate the fused feature map F′: F′=FPN(F) Color distortion correction: Correct the color distortion caused by the light source through the color restoration algorithm. Assume that the color of the input image is C in , the corrected color is C out ,but: C out =CR(C in ) Among them, CR represents the color restoration algorithm; S4, color evaluation and color difference calculation; Design a color evaluation algorithm and use Lab value and ΔE value to accurately evaluate the color. Lab value calculation: Assuming that the RGB value in the image is (R, G, B), its corresponding Lab value is (L, a, b), which can be calculated using the following conversion formula: b=200[f(Y / Y n )-f(Z / Z n )] Among them, X n , Y n , Z n is the XYZ value of the reference white point, ΔE value calculation: ΔE value is a standard used to measure the difference between two colors. The calculation formula is: By calculating the ΔE value, the color error is accurately evaluated and the color accuracy is judged according to the set threshold; S5. Standard database construction and optimization; S6, automated operation process; S7. Model optimization and real-time data update.

2. The high-precision color recognition method for complex light source environments according to claim 1, wherein: The hardware system in S1 uses an area array camera, liquid lens, and laser camera, and provides uniform and stable illumination through a parallel light source; Area array cameras are used to acquire high-resolution image data. Liquid lenses can adjust the focal length according to the size and position of different objects to ensure image clarity. Laser cameras provide additional depth information to enhance the ability to capture image details. Parallel light sources eliminate the impact of light source unevenness on the image through uniform lighting conditions, thereby ensuring the authenticity of color reproduction.

3. The high-precision color recognition method for complex light source environments according to claim 1, wherein: In the construction and optimization of the standard database in S5, a standard database is built to store standard color data under different light sources. The data in the database is calibrated and verified to ensure its accuracy. By learning and optimizing historical data, the color recognition accuracy of new samples is continuously improved. The construction process of the standard database includes: 1) Collect standard color data under different light sources; 2) Calibrate and verify the data to ensure its accuracy; 3) Use machine learning algorithms to optimize data and improve the quality and reliability of the database.

4. The high-precision color recognition method for complex light source environments according to claim 1, wherein: The automated operation process in S6 realizes full automation of the entire process from sample entry to final result display. Users can complete all tasks through simple interface settings. It automatically completes image acquisition, processing, feature extraction, color correction, color evaluation, and finally outputs a detailed identification report. The specific process is as follows: 1) The user enters sample information; 2) The system automatically collects images and performs preprocessing; 3) The system performs feature extraction, color correction and color evaluation on the image; 4) Output the results and generate a color difference report.

5. The high-precision color recognition method for complex light source environments according to claim 1, wherein: The specific optimization process of model optimization and real-time data update in S7 is as follows: Model training: Based on the enhanced training data set, the deep learning model is used for training. The improved YOLO model is selected to obtain the feature map F of the sample image. The feature map is processed using the improved feature pyramid network and then feature fusion F′ is performed. The detection head is used to analyze the feature map and predict the target category and position coordinates of each detection frame. The detection head also predicts the confidence P of each detection frame, which indicates the probability that the detection frame contains the target. A confidence threshold T is set. Only detection frames with a confidence higher than the threshold will be judged as having defects. If P>T, it is determined that there is a defect Cross-validation: During the training process, the model is trained and validated multiple times using the cross-validation method to ensure the generalization ability of the model; Performance evaluation: Use the test data set to test the trained model and evaluate the performance of the model. By calculating the accuracy A, recall R, and F1 value F1, the detection effect of the model is comprehensively evaluated; Among them, TP represents true positive examples, FP represents false positive examples, and FN represents false negative examples.

Images