Image quality evaluation method and system
By constructing a convolutional neural network model and using bilinear interpolation and softmax normalization methods, the problem of insufficient applicability and generalization capabilities in image quality evaluation is solved, and accurate evaluation and rapid feedback are achieved under complex conditions.
Patent Information
- Application Number
- CN202510136308.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-07
- Publication Date
- 2025-06-06
AI Technical Summary
The prior art has problems such as insufficient applicability and generalization capabilities, strong dependence on labeled data, limitations on reference image requirements, and poor robustness of evaluation results in image quality evaluation.
An image quality evaluation method and system are provided. By constructing an image quality evaluation model including the first convolutional neural network and the second convolutional neural network, using bilinear interpolation and softmax normalization methods, the appropriate evaluation strategy is automatically selected to adapt to different imaging conditions.
It achieves the accuracy and consistency of image quality evaluation under complex conditions, avoids the inaccurate evaluation of traditional models, and has the characteristics of fast feedback and high evaluation accuracy, which is suitable for edge computing or real-time image processing requirements.
Smart Images
Figure CN120107172A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of computer vision technology, and specifically provides an image quality assessment method and system. Background Art
[0002] Accurate assessment of image quality is a critical task in fields such as image processing, computer vision, and remote perception. High-quality image assessment can be used to identify problems such as noise, blur, and poor contrast in images, ensuring that the image achieves the desired effect in subsequent processing.
[0003] Traditional image quality assessment methods usually rely on subjective manual evaluation or algorithms based on indicators such as contrast, clarity, brightness, etc. However, these methods often perform poorly when faced with complex, dynamic, or uneven image distributions. With the widespread application of various imaging technologies, image quality assessment is no longer limited to a single scene or static environment, but needs to process images under different imaging conditions (such as illumination changes, noise interference, multiple devices, etc.). Therefore, how to ensure the accuracy and consistency of evaluation results under these different conditions has become an important technical issue.
[0004] AI-based evaluation methods use machine learning models and a large amount of labeled data to train image quality evaluation models. AI-based evaluation methods can improve the accuracy of image quality evaluation to a certain extent, especially when there is abundant labeled data. However, such methods are highly data-dependent and usually require large-scale labeled image sets to support model training. In addition, these methods have limited generalization capabilities and are not stable enough when faced with new scenarios or different data distributions.
[0005] In the prior art, there is a non-reference statistical method, that is, in the absence of lossless reference images, some non-reference quality assessment methods try to use image statistical features to infer quality. For example, based on the statistical features of noise, the distribution of texture information, etc. These methods evaluate quality by calculating whether the feature distribution of the image meets certain statistical features, but it is difficult to accurately adapt to different imaging conditions, and the robustness and accuracy of the model are poor in complex scenes.
[0006] The shortcomings of existing technologies are insufficient applicability and generalization ability, strong dependence on labeled data, limited reference image requirements, and poor robustness of evaluation results. Specifically, manual feature extraction methods are usually designed based on specific scenarios and are difficult to adapt to complex and changing imaging conditions; artificial intelligence-based evaluation methods require a large amount of high-quality labeled data support, and when data is insufficient or encounters new scenarios, they are unstable and have limited generalization ability; reference image evaluation methods rely on lossless reference images as the "gold standard", but it is often difficult to obtain suitable reference images in practical applications, which limits their scope of application; and statistical methods without references lack robustness and consistency when processing images with different lighting, noise or equipment differences, making it difficult to ensure accurate quality evaluation. Summary of the invention
[0007] In order to solve the above problems, the present invention provides an image quality assessment method and system, which can automatically select a suitable assessment strategy according to the characteristics of the image during the assessment process, thereby avoiding the problem of inaccurate assessment of traditional models under complex conditions.
[0008] In a first aspect, the present invention provides an image quality assessment method, comprising: The quality score of the data to be evaluated in the data set is evaluated in advance to complete the scoring of the image quality, and the training data set is obtained after the labeling is completed; Constructing an image quality assessment model, the image quality assessment model comprising a first convolutional neural network and a second convolutional neural network, the first convolutional neural network being used to learn the noise type of the image, the second convolutional neural network being used to learn the noise level of the image, performing bilinear interpolation on the prediction results of the first convolutional neural network and the second convolutional neural network to obtain an image quality assessment result based on bilinear interpolation, training the image quality assessment model using the training data set until convergence to obtain a trained image quality assessment model, and saving the trained weight parameters; The image to be evaluated is obtained and input into the trained image quality evaluation model to obtain an image quality evaluation result of the image to be evaluated.
[0009] As a preferred solution, the quality score of the data to be evaluated in the data set is evaluated in advance to complete the scoring of the image quality, and the training data set is obtained after the labeling is completed, including: Log in the labeler information, select the image file path to be evaluated, and have the labeler evaluate the quality score of the data to be evaluated in the data set. According to subjective consistency, score the quality of the image. The score is divided into 5 standards, excellent, good, medium, poor, and poor. The results are stored in the database file according to the labeler number, file path, score and labeling time to complete the construction of the training data set.
[0010] As a preferred solution, it also includes: The image quality assessment model also includes a fused convolutional neural network, which is used to perform a bilinear interpolation operation on the prediction results of the first convolutional neural network and the second convolutional neural network, and then obtain the image quality assessment result through a softmax normalization method.
[0011] In a second aspect, the present invention provides an image quality assessment system, comprising: The labeling unit is used to pre-evaluate the quality score of the data to be evaluated in the data set to complete the scoring of the image quality, and obtain the training data set after the labeling is completed; A model training unit, used to construct an image quality assessment model, the image quality assessment model includes a first convolutional neural network and a second convolutional neural network, the first convolutional neural network is used to learn the noise type of the image, the second convolutional neural network is used to learn the noise level of the image, based on bilinear interpolation, bilinear interpolation is performed on the prediction results of the first convolutional neural network and the second convolutional neural network to obtain an image quality assessment result, the image quality assessment model is trained using the training data set until convergence to obtain a trained image quality assessment model, and the trained weight parameters are saved; The image quality assessment unit is used to obtain the image to be assessed and input the image into the trained image quality assessment model to obtain the image quality assessment result of the image to be assessed.
[0012] As a preferred solution, the labeling unit is specifically used for: Log in the labeler information, select the image file path to be evaluated, and have the labeler evaluate the quality score of the data to be evaluated in the data set. According to subjective consistency, score the quality of the image. The score is divided into 5 standards, excellent, good, medium, poor, and poor. The results are stored in the database file according to the labeler number, file path, score and labeling time to complete the construction of the training data set.
[0013] As a preferred solution, it also includes: The image quality assessment model also includes a fused convolutional neural network, which is used to perform a bilinear interpolation operation on the prediction results of the first convolutional neural network and the second convolutional neural network, and then obtain the image quality assessment result through a softmax normalization method.
[0014] Compared with the prior art, the present invention can achieve the following beneficial effects: The present invention provides an image quality assessment method and system, which pre-evaluates the quality score of the data to be evaluated in the data set to complete the scoring of the image quality, obtains the training data set after the annotation, and constructs an image quality assessment model, wherein the image quality assessment model includes a first convolutional neural network and a second convolutional neural network, wherein the first convolutional neural network is used to learn the noise type of the image, and the second convolutional neural network is used to learn the noise level of the image, and based on bilinear interpolation, the prediction results of the first convolutional neural network and the second convolutional neural network are bilinearly interpolated to obtain the image quality assessment result, and the image quality assessment model is trained by using the training data set until convergence to obtain a trained image quality assessment model, and the trained weight parameters are saved, and the image to be evaluated is obtained, and the image quality assessment result of the image to be evaluated is obtained. By using an efficient computing strategy, quality assessment feedback is quickly provided after the image is input, which is particularly suitable for edge computing or real-time image processing requirements. By introducing a computing acceleration method and a memory optimization strategy, high evaluation accuracy and fast feedback can be maintained under limited computing resources to meet the requirements of real-time and accuracy in practical applications. BRIEF DESCRIPTION OF THE DRAWINGS
[0015] Figure 1 is a flow chart of an image quality assessment method provided according to an embodiment of the present invention; Figure 2 is a structural block diagram of an image quality assessment system provided according to an embodiment of the present invention. DETAILED DESCRIPTION
[0016] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. In the following description, the same modules are represented by the same reference numerals. In the case of the same reference numerals, their names and functions are also the same. Therefore, the detailed description thereof will not be repeated.
[0017] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and do not constitute a limitation of the present invention.
[0018] Combination Figure 1 As shown, an embodiment of the present invention provides an image quality assessment method, comprising: S101, pre-evaluate the quality score of the data to be evaluated in the data set to complete the scoring of the image quality, and obtain the training data set after completing the labeling.
[0019] S102. Construct an image quality assessment model, where the image quality assessment model includes a first convolutional neural network and a second convolutional neural network, where the first convolutional neural network is used to learn the noise type of the image, and the second convolutional neural network is used to learn the noise level of the image. Based on bilinear interpolation, bilinear interpolation is performed on the prediction results of the first convolutional neural network and the second convolutional neural network to obtain an image quality assessment result. The image quality assessment model is trained using the training data set until convergence to obtain a trained image quality assessment model, and the trained weight parameters are saved.
[0020] S103: Acquire an image to be evaluated, and input the image into the trained image quality evaluation model to obtain an image quality evaluation result of the image to be evaluated.
[0021] An image quality assessment method provided in an embodiment of the present invention is to pre-evaluate the quality score of the data to be evaluated in the data set to complete the scoring of the image quality, obtain the training data set after the annotation is completed, and construct an image quality assessment model, wherein the image quality assessment model includes a first convolutional neural network and a second convolutional neural network, wherein the first convolutional neural network is used to learn the noise type of the image, and the second convolutional neural network is used to learn the noise level of the image, and based on bilinear interpolation, the prediction results of the first convolutional neural network and the second convolutional neural network are bilinearly interpolated to obtain the image quality assessment result, and the image quality assessment model is trained by using the training data set until convergence to obtain a trained image quality assessment model, and the trained weight parameters are saved, and the image to be evaluated is obtained, and the image quality assessment result of the image to be evaluated is obtained. By using an efficient computing strategy, quality assessment feedback is quickly provided after the image is input, which is particularly suitable for edge computing or real-time image processing requirements. By introducing a computing acceleration method and a memory optimization strategy, high evaluation accuracy and fast feedback can be maintained under limited computing resources to meet the requirements of real-time and accuracy in practical applications.
[0022] In step S101, the quality score of the data to be evaluated in the data set is evaluated in advance to complete the scoring of the image quality, and the training data set is obtained after the labeling is completed, including: Log in the labeler information, select the image file path to be evaluated, and have the labeler evaluate the quality score of the data to be evaluated in the dataset. According to subjective consistency, the quality of the image is scored. The scores are divided into 5 standards: excellent, good, medium, poor, and bad. The results are stored in the database file according to the labeler number, file path, score, and labeling time to complete the construction of the training dataset.
[0023] It should be noted that the scoring criteria for image quality can be flexibly set as needed and there is no limitation on this.
[0024] As a preferred solution, it also includes: The image quality assessment model also includes a fused convolutional neural network, which is used to perform a bilinear interpolation operation on the prediction results of the first convolutional neural network and the second convolutional neural network, and then obtain the image quality assessment result through a softmax normalization method.
[0025] The image quality assessment model designs two convolutional neural networks to learn the noise type and noise level of the image respectively, and fuses the results through bilinear interpolation to generate the final image quality evaluation result.
[0026] In this embodiment, the specific network structure of the image quality assessment model can be seen in Table 1.
[0027] Table 1 Network structure
[0028] The above table shows the network structure of the image quality assessment model, where the first convolutional neural network NET1 is responsible for predicting the noise type of the image, and the second convolutional neural network NET2 is responsible for predicting the noise level of the image. The prediction results of the first convolutional neural network NET1 and the second convolutional neural network NET2 are bilinearly interpolated by fusing the convolutional neural network UNION_NET, and then the softmax method is used to obtain the final image quality score.
[0029] The image quality assessment method provided by the present invention realizes dynamic assessment of images to adjust the calculation strategy to adapt to imaging conditions such as different lighting, equipment differences and noise interference, thereby ensuring the stability and consistency of the assessment results. During the assessment process, the model can automatically select the appropriate assessment strategy according to the characteristics of the image, thus avoiding the problem of inaccurate assessment of traditional models under complex conditions.
[0030] Combined accordingly, Figure 2 As shown, an embodiment of the present invention further provides an image quality assessment system, comprising: The labeling unit 201 is used to pre-evaluate the quality score of the data to be evaluated in the data set to complete the scoring of the image quality, and obtain the training data set after the labeling is completed; A model training unit 202 is used to construct an image quality assessment model, wherein the image quality assessment model includes a first convolutional neural network and a second convolutional neural network, wherein the first convolutional neural network is used to learn the noise type of the image, and the second convolutional neural network is used to learn the noise level of the image, and based on bilinear interpolation, bilinear interpolation is performed on the prediction results of the first convolutional neural network and the second convolutional neural network to obtain an image quality assessment result, and the image quality assessment model is trained using the training data set until convergence to obtain a trained image quality assessment model, and the trained weight parameters are saved; The image quality assessment unit 203 is used to obtain the image to be assessed and input the image into the trained image quality assessment model to obtain the image quality assessment result of the image to be assessed.
[0031] As a preferred solution, the labeling unit 201 is specifically used for: Log in the labeler information, select the image file path to be evaluated, and have the labeler evaluate the quality score of the data to be evaluated in the data set. According to subjective consistency, score the quality of the image. The score is divided into 5 standards, excellent, good, medium, poor, and poor. The results are stored in the database file according to the labeler number, file path, score and labeling time to complete the construction of the training data set.
[0032] As a preferred solution, it also includes: The image quality assessment model also includes a fused convolutional neural network, which is used to perform a bilinear interpolation operation on the prediction results of the first convolutional neural network and the second convolutional neural network, and then obtain the image quality assessment result through a softmax normalization method.
[0033] The embodiment of the present invention also provides an image quality assessment system, which uses an efficient computing strategy to quickly provide quality assessment feedback after the image is input, and is particularly suitable for edge computing or real-time image processing needs. By introducing computing acceleration methods and memory optimization strategies, high assessment accuracy and fast feedback can be maintained under limited computing resources to meet the requirements of real-time and accuracy in practical applications.
[0034] Although the embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and cannot be understood as limiting the present invention. Those skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.
[0035] The above specific implementations of the present invention do not constitute a limitation on the protection scope of the present invention. Any other corresponding changes and modifications made based on the technical concept of the present invention should be included in the protection scope of the claims of the present invention.
Claims
1. A method for evaluating image quality, characterized in that: include: The quality score of the data to be evaluated in the data set is evaluated in advance to complete the scoring of the image quality, and the training data set is obtained after the labeling is completed; Constructing an image quality assessment model, the image quality assessment model comprising a first convolutional neural network and a second convolutional neural network, the first convolutional neural network being used to learn the noise type of the image, the second convolutional neural network being used to learn the noise level of the image, performing bilinear interpolation on the prediction results of the first convolutional neural network and the second convolutional neural network to obtain an image quality assessment result based on bilinear interpolation, training the image quality assessment model using the training data set until convergence to obtain a trained image quality assessment model, and saving the trained weight parameters; The image to be evaluated is obtained and input into the trained image quality evaluation model to obtain an image quality evaluation result of the image to be evaluated.
2. The image quality assessment method according to claim 1, characterized in that: The step of pre-evaluating the quality score of the data to be evaluated in the data set to complete the scoring of the image quality, and obtaining the training data set after completing the labeling, includes: Log in the labeler information, select the image file path to be evaluated, and have the labeler evaluate the quality score of the data to be evaluated in the data set. According to subjective consistency, score the quality of the image. The score is divided into 5 standards, excellent, good, medium, poor, and poor. The results are stored in the database file according to the labeler number, file path, score and labeling time to complete the construction of the training data set.
3. The image quality assessment method according to claim 1, wherein: Also includes: The image quality assessment model also includes a fused convolutional neural network, which is used to perform a bilinear interpolation operation on the prediction results of the first convolutional neural network and the second convolutional neural network, and then obtain the image quality assessment result through a softmax normalization method.
4. An image quality assessment system, characterized in that: include: The labeling unit is used to pre-evaluate the quality score of the data to be evaluated in the data set to complete the scoring of the image quality, and obtain the training data set after the labeling is completed; A model training unit, used to construct an image quality assessment model, the image quality assessment model includes a first convolutional neural network and a second convolutional neural network, the first convolutional neural network is used to learn the noise type of the image, the second convolutional neural network is used to learn the noise level of the image, based on bilinear interpolation, bilinear interpolation is performed on the prediction results of the first convolutional neural network and the second convolutional neural network to obtain an image quality assessment result, the image quality assessment model is trained using the training data set until convergence to obtain a trained image quality assessment model, and the trained weight parameters are saved; The image quality assessment unit is used to obtain the image to be assessed and input the image into the trained image quality assessment model to obtain the image quality assessment result of the image to be assessed.
5. The image quality assessment system according to claim 4, characterized in that: The marking unit is specifically used for: Log in the labeler information, select the image file path to be evaluated, and have the labeler evaluate the quality score of the data to be evaluated in the data set. According to subjective consistency, score the quality of the image. The score is divided into 5 standards, excellent, good, medium, poor, and poor. The results are stored in the database file according to the labeler number, file path, score and labeling time to complete the construction of the training data set.
6. The image quality assessment system according to claim 4, characterized in that: Also includes: The image quality assessment model also includes a fused convolutional neural network, which is used to perform a bilinear interpolation operation on the prediction results of the first convolutional neural network and the second convolutional neural network, and then obtain the image quality assessment result through a softmax normalization method.