Image quality evaluation method, device, storage medium, and electronic device

Abstract

The present disclosure provides an image quality evaluation method, device, storage medium, and electronic device, including: obtaining an image to be evaluated and a generated keyword corresponding to the image to be evaluated; the image to be evaluated is an image generated by a trained image model through the generated keyword; for the image to be evaluated, using an evaluation algorithm corresponding to at least one mechanical evaluation index, and calculating the matching result between the image to be evaluated and the generated keyword; inputting the image to be evaluated and the generated keyword into a preset simulated manual evaluation module, and evaluating the image to be evaluated from multiple preset evaluation dimensions through the simulated manual evaluation module to obtain the evaluation results corresponding to each evaluation dimension; finding the target quality evaluation corresponding to the matching result and the evaluation result, and determining the target quality evaluation as the quality corresponding to the image to be evaluated. The present disclosure establishes an evaluation dimension to evaluate the quality of AI images from the perspective of user perception on the basis of mechanical evaluation indicators, so that the evaluation results are more accurate and reliable.

Description

Technical Field

[0001] The present disclosure relates to the field of AI image evaluation technology, and in particular to an image quality evaluation method, device, storage medium, and electronic device. Background Art

[0002] As the field of AI imaging becomes more and more mature, users can generate the desired AI images by inputting keywords into various AI imaging models. However, the quality of AI images generated by different AI imaging models varies. Therefore, in order to facilitate users to select AI imaging models based on image quality, it is urgent to evaluate the quality of AI images.

[0003] Currently, mechanical evaluation indicators (such as FID, IS, and Clip Score) are usually used to evaluate the quality of AI images. Mechanical evaluation indicators evaluate image quality from a computer's perspective. Mechanical evaluation indicators may give evaluations that are very different from human perception, and often produce results that are different from the "high quality" perceived by users. For example, the mechanical evaluation indicators of an AI image are high, but the user's perception may be that the quality of the AI ​​image is not high. Summary of the invention

[0004] In order to solve the above problems, the present disclosure provides an image quality assessment method, device, storage medium, electronic device, and image quality assessment vehicle, which solves the problem that the quality assessment of AI images based on mechanical evaluation indicators is different from the high quality perceived by users, and optimizes the quality assessment method of AI images.

[0005] According to a first aspect of the present disclosure, the present disclosure provides a method for evaluating image quality, comprising:

[0006] Acquire an image to be evaluated and a generated keyword corresponding to the image to be evaluated; the image to be evaluated is an image generated based on the generated keyword according to a preset text-generated image model;

[0007] For the image to be evaluated, using an evaluation algorithm corresponding to at least one mechanical evaluation index to calculate a matching result between the image to be evaluated and the generated keyword;

[0008] Inputting the image to be evaluated and the generated keywords into a preset simulated manual evaluation module, and evaluating the image to be evaluated from at least one preset user-perceived evaluation dimension through the simulated manual evaluation module to obtain an evaluation result corresponding to the evaluation dimension;

[0009] The preset quality evaluation level corresponding to the matching result is searched, and the preset quality level corresponding to the evaluation result is searched, and the preset quality level with the largest proportion is determined as the quality evaluation result corresponding to the image to be evaluated.

[0010] As an optional implementation of the embodiment of the present disclosure, the using of an evaluation algorithm corresponding to at least one mechanical evaluation index to calculate a matching result between the image to be evaluated and the generated keyword includes:

[0011] Performing feature vector conversion on the image to be evaluated and the generated keywords respectively to obtain an image feature vector corresponding to the image to be evaluated and a text feature vector corresponding to the generated keywords;

[0012] The similarity between the image feature vector and the text feature vector is calculated, and the similarity is used as the matching result.

[0013] As an optional implementation of the embodiment of the present disclosure, the evaluation dimension includes: at least one of content correctness, content relevance, and aesthetics;

[0014] The step of evaluating the image to be evaluated from at least one preset user-perceived evaluation dimension by the simulated manual evaluation module includes:

[0015] When the evaluation dimension includes the content correctness, searching whether there is an object corresponding to the generated keyword in the image to be evaluated, and determining the coverage of the generated keyword by the image to be evaluated;

[0016] When the evaluation dimension includes the content relevance, comparing the generated keywords with the morphological parameters of each object in the image to be evaluated, and determining the number of objects with incorrect morphology in the image to be evaluated;

[0017] When the evaluation dimension includes the aesthetics, it is determined whether the style of the image to be evaluated is consistent with the style in the generated keywords to obtain a corresponding style consistency result, and the aesthetics of the image to be evaluated is scored to obtain a corresponding aesthetics score.

[0018] As an optional implementation of the embodiment of the present disclosure, when the evaluation dimension includes the content correctness, searching whether there is an object corresponding to the generated keyword in the image to be evaluated, and determining the coverage of the generated keyword by the image to be evaluated includes:

[0019] In view of the correctness of the content, performing an image-to-text operation on the image to be evaluated by using a preset image-to-text algorithm to obtain a text description corresponding to the image to be evaluated;

[0020] The similarity between the text description and the generated keywords is calculated, and based on the similarity, the coverage of the image to be evaluated for the generated keywords is determined, and the coverage is determined as the evaluation result corresponding to the content correctness.

[0021] As an optional implementation of the embodiment of the present disclosure, when the evaluation dimension includes the content relevance, comparing the generated keywords with the morphological parameters of each object in the image to be evaluated to determine the number of objects with incorrect morphology in the image to be evaluated includes:

[0022] According to the content relevance, a preset image recognition algorithm is used to perform object recognition on the image to be evaluated, so as to obtain morphological parameters of each object in the image to be evaluated;

[0023] The preset morphological parameters of each object in the generated keyword are compared with the morphological parameters to determine the number of objects with incorrect morphology in the image to be evaluated, and the number of objects with incorrect morphology is determined as the evaluation result corresponding to the content relevance.

[0024] As an optional implementation of the embodiment of the present disclosure, when the evaluation dimension includes the aesthetics, determining whether the style of the image to be evaluated is consistent with the style in the generated keywords, obtaining a corresponding style consistency result, and scoring the aesthetics of the image to be evaluated to obtain a corresponding aesthetics score includes:

[0025] With respect to the aesthetics, the style of the image to be evaluated is predicted by a style prediction algorithm, and it is determined whether the predicted style is consistent with the style in the generated keywords, thereby obtaining a corresponding style consistency result;

[0026] Using a preset aesthetic scoring algorithm, the image to be evaluated is scored for aesthetics to obtain a corresponding aesthetic score;

[0027] The style consistency result and the aesthetic score are determined as the evaluation result corresponding to the aesthetic.

[0028] As an optional implementation of the embodiment of the present disclosure, the simulated manual evaluation module is a preset visual question answering model; the step of evaluating the image to be evaluated from at least one preset user-perceived evaluation dimension by the simulated manual evaluation module includes:

[0029] Performing image recognition on the image to be evaluated by using the visual question answering model to obtain answers to a plurality of preset manual evaluation related questions;

[0030] Reference answers corresponding to the multiple manual evaluation related questions are extracted from the generated keywords, the reference answers are compared with the preset answers to the multiple manual evaluation related questions to obtain multiple comparison results, the multiple comparison results are mapped to at least one preset evaluation dimension, and the evaluation results corresponding to the at least one preset evaluation dimension are obtained.

[0031] As an optional implementation of the embodiment of the present disclosure, mapping the multiple comparison results to at least one preset evaluation dimension to obtain the evaluation result corresponding to the at least one preset evaluation dimension includes:

[0032] Classify multiple comparison results according to multiple preset evaluation dimensions;

[0033] For each evaluation dimension, each comparison result classified into the evaluation dimension is multiplied and summed with the corresponding weight coefficient to obtain the evaluation result corresponding to the evaluation dimension.

[0034] According to a second aspect of the present disclosure, the present disclosure provides an image quality assessment device, comprising:

[0035] An acquisition module, used for acquiring an image to be evaluated and a generated keyword corresponding to the image to be evaluated; the image to be evaluated is an image generated based on the generated keyword according to a preset text-generated image model;

[0036] A calculation module, configured to calculate a matching result between the image to be evaluated and the generated keyword by using an evaluation algorithm corresponding to at least one mechanical evaluation index for the image to be evaluated;

[0037] An evaluation module, configured to input the image to be evaluated and the generated keywords into a preset simulated manual evaluation module, and evaluate the image to be evaluated from at least one preset user-perceived evaluation dimension through the simulated manual evaluation module to obtain an evaluation result corresponding to the evaluation dimension;

[0038] The determination module is used to find the preset quality evaluation level corresponding to the matching result and the preset quality level corresponding to the evaluation result, and determine the preset quality level with the largest proportion as the quality evaluation result corresponding to the image to be evaluated.

[0039] According to a third aspect of the present disclosure, the present disclosure provides a readable storage medium having a computer program stored thereon, characterized in that when the computer program is executed by a processor, the image quality assessment method as described in any one of the embodiments of the first aspect is implemented.

[0040] According to a fourth aspect of the present disclosure, the present disclosure provides an electronic device, including:

[0041] at least one processor; and

[0042] a memory communicatively connected to the at least one processor; wherein,

[0043] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can implement the steps of the image quality assessment method as described in any embodiment of the first aspect.

[0044] According to a fifth aspect of the present disclosure, the present disclosure provides a vehicle, comprising: an image quality assessment device as described in the second aspect above.

[0045] The technical solution provided by the embodiment of the present disclosure has the following advantages compared with the prior art: the present disclosure provides a method for evaluating the quality of an image, including: obtaining an image to be evaluated and a generated keyword corresponding to the image to be evaluated; the image to be evaluated is an image generated based on the generated keyword according to a preset text-based image model; for the image to be evaluated, using an evaluation algorithm corresponding to at least one mechanical evaluation index to calculate the matching result between the image to be evaluated and the generated keyword; inputting the image to be evaluated and the generated keyword into a preset simulated manual evaluation module, evaluating the image to be evaluated from at least one preset user-perceived evaluation dimension through the simulated manual evaluation module, and obtaining the evaluation result corresponding to the evaluation dimension; searching for the preset quality evaluation level corresponding to the matching result, and searching for the preset quality level corresponding to the evaluation result, and determining the preset quality level with the largest proportion as the quality evaluation result corresponding to the image to be evaluated. The present disclosure establishes an evaluation dimension to evaluate the quality of AI images from the perspective of user perception on the basis of mechanical evaluation indicators, which not only saves labor costs, but also enables the evaluation results to combine the mechanical dimension and the user-perceived evaluation dimension, which is more accurate and reliable.

[0046] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present application, nor is it intended to limit the scope of the present application. Other features of the present application will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0048] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0049] Figure 1 is a flowchart of an image quality assessment method provided by the first aspect of the embodiment of the present disclosure;

[0050] Figure 2 is another implementation flow diagram of an image quality assessment method provided by the first aspect of the embodiment of the present disclosure;

[0051] Figure 3 is another implementation flow diagram of an image quality assessment method provided by the first aspect of the embodiment of the present disclosure;

[0052] Figure 4 is a schematic diagram of an application framework of an image quality assessment method provided by the first aspect of the embodiment of the present disclosure;

[0053] Figure 5 is a structural schematic diagram of an image quality assessment device provided by the second aspect of an embodiment of the present disclosure;

[0054] Figure 6 It is a structural diagram of an electronic device provided in the fourth aspect of an embodiment of the present disclosure. DETAILED DESCRIPTION

[0055] In the following description, many specific details are set forth to facilitate a full understanding of the present disclosure, but the present disclosure may also be implemented in other ways different from those described herein; it is obvious that the embodiments in the specification are only part of the embodiments of the present disclosure, rather than all of the embodiments.

[0056] In the embodiments of the present application, words such as "exemplary" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a concrete way. In addition, in the description of the embodiments of the present application, unless otherwise specified, the meaning of "multiple" refers to two or more.

[0057] It should be noted that, in this article, the term "comprises" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0058] The present application provides a method for evaluating image quality. Figure 1 is a flow chart of an image quality evaluation method provided by the first aspect of the embodiment of the present disclosure. The evaluation method can be performed by an evaluation device or equipment. The device or equipment can be configured in a server, a processor or a main control chip. For example, it can be an electronic terminal device such as a computer, a mobile phone, or a vehicle system of a vehicle. Figure 1 As shown, the image quality evaluation method includes the following steps S101-S104:

[0059] S101: Obtain an image to be evaluated and generated keywords corresponding to the image to be evaluated.

[0060] This technical solution is applied in the field of generative artificial intelligence AIGC (Artificial Intelligence Generated Content). The image to be evaluated is an image generated based on the generated keywords according to a preset text-generated image model. Specifically, the artificial intelligence will generate the image to be evaluated using the AI ​​image model through the running image interface according to the generated keyword prompt input by the user.

[0061] As an optional embodiment of the present disclosure, the image to be evaluated generated by the running image interface and the generated keyword prompt input by the user corresponding to the image to be evaluated are obtained.

[0062] S102 : For the image to be evaluated, using an evaluation algorithm corresponding to at least one mechanical evaluation index, to calculate a matching result between the image to be evaluated and the generated keyword.

[0063] As an optional embodiment of the present disclosure, the evaluation algorithm corresponding to at least one mechanical evaluation indicator is used to calculate the matching result between the image to be evaluated and the generated keyword, including: performing feature vector conversion on the image to be evaluated and the generated keyword respectively to obtain the image feature vector corresponding to the image to be evaluated and the text feature vector corresponding to the generated keyword; calculating the similarity between the image feature vector and the text feature vector, and taking the similarity as the matching result.

[0064] As an optional embodiment of the present disclosure, the similarity between the image feature vector and the text feature vector is calculated. For the image to be evaluated, one or more evaluation algorithms corresponding to mechanical evaluation indicators can be used. Specifically, the mechanical evaluation indicators include but are not limited to FID (Frechet Inception Distance score, FID), CLIPScore, VSS (Visual Semantic Similarity), IS (Inception Score), etc. The image to be evaluated and the generated keywords are converted into an image feature vector and a text feature vector using the Inception algorithm. For the FID indicator, it is assumed that the distribution of the two feature vectors is a Gaussian distribution, and their means and covariance matrices are calculated. The similarity between the image feature vector and the text feature vector is evaluated by measuring the distance between the probability distributions, that is, the degree of similarity, and the similarity is used as the matching result. When the FID value is smaller, the similarity is greater, the matching result is more ideal, and the image quality is higher. For the CLIP Score indicator, the cosine similarity between the image feature vector and the text feature vector is calculated, and the cosine similarity is used as the matching result. When the CLIP The larger the Score value, the higher the matching degree between the image and the text, and the better the image quality. For the VSS index, the image feature vector and the text feature vector are mapped to a common semantic space, and the cosine similarity in a common semantic space is calculated. When the VSS value is larger, the matching degree between the image and the text is higher, and the image quality is better. For the IS index, the information entropy of the image is calculated with the help of an image classifier. If the image has a good confidence level and is judged to belong to a certain category, it means that the image has good clarity and recognition. If the information entropy value of the probability distribution of different categories is larger, it means that the image is more diverse. IS combines the two values, and calculates the expectation of KL divergence through marginal distribution and conditional distribution to obtain the value of the IS index. When the IS value is larger, the image quality is better. By evaluating the image to be evaluated through mechanical evaluation indicators, the objective data of the image to be evaluated can be used to present the quality of the image.

[0065] S103, inputting the image to be evaluated and the generated keywords into a preset simulated manual evaluation module, and evaluating the image to be evaluated from at least one preset user-perceived evaluation dimension through the simulated manual evaluation module to obtain an evaluation result corresponding to the evaluation dimension.

[0066] As an optional embodiment of the present disclosure, the image to be evaluated is evaluated according to relevant indicators of three dimensions: content correctness, content relevance, and content aesthetics, and evaluation results corresponding to each evaluation dimension are obtained. Specifically, the image to be evaluated in three dimensions: content correctness, content relevance, and content aesthetics are selected from the perspective of user perception for evaluation. With respect to the content correctness, it is found whether each word in the generated keywords exists in the image to be evaluated, and the coverage rate of the generated keywords by the image to be evaluated is determined. Specifically, a preset image-to-text algorithm is used to perform image-to-text operations on the image to be evaluated, and a text description corresponding to the image to be evaluated is obtained. The similarity between the text description and the generated keywords is calculated, and the coverage rate of the image to be evaluated for the generated keywords is determined based on the similarity, and the coverage rate is determined as the evaluation result corresponding to the content correctness. With respect to the content relevance, the generated keywords and the morphological parameters of each object in the image to be evaluated are compared to determine the number of objects with incorrect morphology in the image to be evaluated, wherein the objects include tangible objects such as people, plants, food, and animals. Specifically, the preset image recognition algorithm is used to perform object recognition on the image to be evaluated to obtain the morphological parameters of each object in the image to be evaluated, the preset morphological parameters of each object in the generated keyword are compared with the morphological parameters, the number of objects with incorrect morphology in the image to be evaluated is determined, and the number of objects with incorrect morphology is determined as the evaluation result corresponding to the content relevance; for the aesthetics, it is determined whether the style of the image to be evaluated is consistent with the style in the generated keyword to obtain the corresponding style consistency result, and the image to be evaluated is scored for aesthetics to obtain the corresponding aesthetics score. Specifically, the style of the image to be evaluated is predicted by a style prediction algorithm, and it is determined whether the predicted style is consistent with the style in the generated keyword to obtain the corresponding style consistency result. The preset aesthetics scoring algorithm is used to score the image to be evaluated for aesthetics to obtain the corresponding aesthetics score. The style consistency result and the aesthetics score are determined as the evaluation result corresponding to the aesthetics. At least one of the above-mentioned user-perceived evaluation dimensions, i.e., any one or a combination of content relevance, content correctness, and aesthetics, can be selected to evaluate the image to be evaluated.

[0067] As an optional embodiment of the present disclosure, the image to be evaluated is subjected to image recognition by the visual question answering model to obtain answers to a plurality of preset manual evaluation related questions; reference answers corresponding to the plurality of manual evaluation related questions are extracted from the generated keywords, the reference answers are compared with the answers to the plurality of preset manual evaluation related questions to obtain a plurality of comparison results, the plurality of comparison results are mapped to at least one preset evaluation dimension to obtain an evaluation result corresponding to the at least one preset evaluation dimension. Specifically, the visual question-answering model asks questions to the image to be evaluated based on a plurality of preset manual evaluation-related questions, and obtains the visual question-answering model's answers to the image to be evaluated, wherein the plurality of preset manual evaluation-related questions may be "What objects are in the image to be evaluated?" "What actions appear in the image to be evaluated?" "What prepositions are in the image to be evaluated?" and the like; reference answers corresponding to the plurality of manual evaluation-related questions are extracted from the generated keywords, the reference answers are compared with the answers to the plurality of preset manual evaluation-related questions to obtain a plurality of comparison results, the plurality of comparison results are mapped to at least one preset evaluation dimension, and the evaluation result corresponding to the at least one preset evaluation dimension is obtained.

[0068] S104, searching for a preset quality evaluation level corresponding to the matching result, and searching for a preset quality level corresponding to the evaluation result, and determining the preset quality level with the largest proportion as the quality evaluation result corresponding to the image to be evaluated.

[0069] As an optional embodiment of the present disclosure, the target quality evaluation corresponding to the matching result and the evaluation result can be found in a preset quality score evaluation table. The preset quality score evaluation table is pre-set according to the average quality score of the AI ​​image, and the score and quality can be adjusted according to the needs. For example, when the score of the mechanical evaluation index VSS in the quality score evaluation table is in the range of 0.8-1, the index score is excellent, when it is in the range of 0.6-0.8, the index score is good, when it is in the range of 0.5-0.6, the index score is medium, and when it is in the range of 0-0.5, the index score is poor; in the manual evaluation module, if the coverage rate of the generated keywords is 80%-100%, the content correctness is excellent, and if the coverage rate of the generated keywords is 60%-8 0%, the content correctness is good. If the coverage of generated keywords is 50%-60%, the content correctness is medium. If the coverage of generated keywords is 0%-50%, the content correctness is poor. It is set in this way. The relevance and aesthetics are similar. When the VSS score is 8.9 and the coverage of generated keywords is 98%, the quality of two evaluation indicators is excellent. The preset quality level with the largest proportion is determined as the quality evaluation result corresponding to the image to be evaluated. Specifically, if the indicators determined to be excellent account for 80% of the total evaluation indicators, the quality evaluation of the image to be evaluated is determined to be excellent. This search method and the specific values ​​in the quality score evaluation table can be freely set according to the requirements of the quality of the image to be evaluated. Here is a method that can be implemented.

[0070] The image quality assessment method provided in the embodiment of the present application is specifically as follows: obtaining an image to be evaluated and generated keywords corresponding to the image to be evaluated; the image to be evaluated is an image generated by a trained image model through the generated keywords; for the image to be evaluated, using an evaluation algorithm corresponding to at least one mechanical evaluation indicator, to calculate the matching result between the image to be evaluated and the generated keywords; inputting the image to be evaluated and the generated keywords into a preset simulated manual evaluation module, and evaluating the image to be evaluated from at least one preset user-perceived evaluation dimension through the simulated manual evaluation module to obtain evaluation results corresponding to each evaluation dimension, wherein the multiple evaluation dimensions are evaluation dimensions corresponding to user perception; searching for a target quality evaluation corresponding to the matching result and the evaluation result, and determining the target quality evaluation as the quality corresponding to the image to be evaluated. This embodiment adds three dimensions of content correctness, content relevance and aesthetics to the evaluation of the image to be evaluated on the basis of existing mechanical evaluation indicators, so as to evaluate the image to be evaluated from the perspective of user perception more closely, compensate for the singleness of the mechanical evaluation indicators, add rich multi-dimensional evaluation indicators, and simultaneously use artificial intelligence to establish evaluation dimensions from the perspective of user perception to evaluate the quality of the image, thus saving labor costs and making the evaluation results more reasonable and reliable.

[0071] As an extension and refinement of the above embodiment, refer to Figure 2 As shown, Figure 2 1 is another implementation flow diagram of an image quality evaluation method provided by the first aspect of the embodiment of the present disclosure, the evaluation method comprises the following steps S201-S204:

[0072] S201: Obtain an image to be evaluated and generated keywords corresponding to the image to be evaluated.

[0073] As an optional embodiment of the present disclosure, the specific method for obtaining the image to be evaluated and the generated keywords corresponding to the image to be evaluated has been described in S101 and will not be repeated here. This step can be implemented with reference to the step in S101, and the electronic device obtains the generated image to be evaluated and the generated keywords corresponding to the image to be evaluated.

[0074] S202 : For the image to be evaluated, using an evaluation algorithm corresponding to at least one mechanical evaluation index, to calculate a matching result between the image to be evaluated and the generated keyword.

[0075] As an optional embodiment of the present disclosure, an evaluation algorithm corresponding to at least one mechanical evaluation indicator is used, and a specific method for calculating the matching result between the image to be evaluated and the generated keyword has been described in S102 and will not be repeated here. This step can be implemented with reference to the step in S102, and one or a combination of several of the mechanical evaluation indicators listed in S102 is selected to obtain the matching result between the image to be evaluated and the generated keyword.

[0076] S203, evaluating the image to be evaluated according to relevant indicators of at least one dimension among the three evaluation dimensions of content correctness, content relevance, and content aesthetics, to obtain evaluation results corresponding to the evaluation dimensions.

[0077] As an optional embodiment of the present disclosure, when the evaluation dimension includes the correctness of the content, it is checked whether there is an object corresponding to the generated keyword in the image to be evaluated, and the coverage of the generated keyword by the image to be evaluated is determined, wherein the object includes tangible objects such as people, plants, food, and animals. First, the image to be evaluated is subjected to an image-to-text operation by an image-to-text algorithm to obtain a text description corresponding to the image to be evaluated, wherein the image-to-text algorithm may adopt an encoder-decoder structure, detect relevant elements such as objects, scenes and actions in the image by an image caption algorithm, fill in relevant words in the template, and combine them into a text description, or adopt an algorithm based on a deep learning model, implement encoding of the image by an encoder, and then generate a corresponding text description by a decoder, or use a convolutional neural network CNN (Convolutional Neural Networks, CNN) to extract image features, and use a time recurrent neural network LSTM (Long Short-Term Memory, LSTM) as a decoder to generate a corresponding text description. The text description corresponding to the image to be evaluated can be obtained by the above method, and the similarity between the text description and the generated keywords is calculated. The keyword matching algorithm can be used, for example, to calculate N-gram similarity, and the original sentence is segmented by length N to obtain word segments, that is, all substrings of length N in the original sentence. For S and T in the text description and the generated keywords, the similarity of the two sentences is defined based on the number of shared substrings:

[0078] Similarity=|GN(S)|+|GN(T)|-2*|GN(S)∩GN(T)|

[0079] Among them, GN(S) and GN(T) represent the sets of N-Grams in strings S and T respectively, N is generally 2 or 3, and the closer the distance between the strings, the greater their similarity.

[0080] Alternatively, by calculating the Jaccard similarity, the ratio between the intersection and the union of the sets of the text description and the generated keywords is calculated. The larger the ratio, the greater the similarity between the text description and the generated keywords. The similarity obtained by the above method can determine the coverage of the image to be evaluated for the generated keywords, and the coverage is determined as the evaluation result corresponding to the correctness of the content.

[0081] As an optional embodiment of the present disclosure, based on the content relevance, the generated keywords are compared with the morphological parameters of each object in the image to be evaluated to determine the number of objects with incorrect morphology in the image to be evaluated, wherein the objects include tangible objects such as people, plants, food, and animals. Specifically, a preset image recognition algorithm, such as a convolutional neural network (CNN), a support vector machine (SVM), a K-nearest neighbor algorithm, etc., is used to detect and identify objects in the image to be evaluated, and data information such as the name, confidence, and coordinates of the object is obtained. Calculations are performed based on the data information to obtain the number of objects in the image to be evaluated, the area of ​​the objects, and the number of object morphological errors. The number of objects in the image to be evaluated is compared with the number of objects in the generated keywords to determine whether the area of ​​the object is greater than 80% and whether there are objects with morphological errors. Through the above data analysis, the number of objects with morphological errors in the image to be evaluated is determined, and the number of objects with morphological errors is determined as the evaluation result corresponding to the content relevance. If the number of objects with morphological errors accounts for 0%-10%, the content relevance is high; if the number of objects with morphological errors accounts for 10%-20%, the content relevance is average; and if the number of objects with morphological errors accounts for more than 20%, the content relevance is poor.

[0082] As an optional embodiment of the present disclosure, with respect to the aesthetics of the content, the style of the image to be evaluated is predicted by a style prediction algorithm, and it is determined whether the predicted style is consistent with the style in the generated keywords. The image to be evaluated is subjected to Fourier transform for denoising, and the arrangement, clarity, and contrast of the pixels are extracted. The style prediction result of the image is obtained by inputting into a deep learning model. The style prediction result may be abstract, realistic, etc., wherein the deep model is trained based on the image information data and the corresponding style in the gallery, and a semantic analysis is performed on the text description of the generated keywords to obtain the image style that the keywords want to obtain, and it is determined whether the predicted style is consistent with the style in the generated keywords, and the corresponding style consistency result is obtained as consistent or inconsistent. When the style consistency result is consistent, the aesthetics is better. Secondly, in the evaluation index of the aesthetics dimension, the image to be evaluated is input into the pre-trained aesthetics model at the same time. The aesthetics model in the technical solution adopts an open source aesthetics model, such as the DARN model. The image pair of the image to be evaluated is input into the network, and after Deep The feature extraction of CNN (such as resnet50) is then passed through three fc layers, and then the variance and mean of the final output are calculated. Assuming that each image is scored by many professional aesthetic experts, then according to the central limit theorem, the image beauty score is normally distributed, so the final calculated mean and variance can be regarded as the average beauty score and variance of the image calculated by the model. When the score value output by the aesthetic model is greater than 3.8 points, the aesthetics of the image to be evaluated is good. The style consistency result and the aesthetics score are compared. The score is determined as the evaluation result corresponding to the aesthetics. For example, when the style consistency result is consistent and the aesthetics score is greater than 3.8, the aesthetics evaluation result of the image to be evaluated is excellent; when the style consistency result is consistent and the aesthetics score is less than 3.8 or the style consistency result is inconsistent and the aesthetics score is greater than 3.8, the aesthetics evaluation result is average; when the style consistency result is inconsistent and the aesthetics score is less than 3.8 or the style consistency result is inconsistent and the aesthetics score is less than 3.8, the aesthetics evaluation result is poor.

[0083] Preferably, the scores of the three dimensions of content correctness, content relevance and aesthetics obtained through this embodiment are more in line with the user's perception perspective to evaluate the image to be evaluated, compensate for the singleness of mechanical evaluation indicators, add rich multi-dimensional evaluation indicators, and make the evaluation results more reasonable and reliable.

[0084] S204, searching for a preset quality evaluation level corresponding to the matching result, and searching for a preset quality level corresponding to the evaluation result, and determining the preset quality level with the largest proportion as the quality evaluation result corresponding to the image to be evaluated.

[0085] As an optional embodiment of the present disclosure, the target quality evaluation corresponding to the matching result and the evaluation result can be searched in a preset quality score evaluation table. The specific setting of the quality score evaluation table has been explained in step S104. The quality score evaluation table method described in step S104 is used for setting. The target quality evaluation corresponding to the matching result and the evaluation result is searched according to the matching result between the image to be evaluated and the generated keyword calculated in steps S202 and S203 and the evaluation results corresponding to the three dimensions of content correctness, content relevance, and aesthetics. The target quality evaluation is determined as the quality corresponding to the image to be evaluated.

[0086] As an extension and refinement of the above embodiment, refer to Figure 3 As shown, Figure 3 1 is another implementation flow diagram of an image quality evaluation method provided by the first aspect of the embodiment of the present disclosure, the evaluation method comprises the following steps S301-S304:

[0087] S301: Obtain an image to be evaluated and generated keywords corresponding to the image to be evaluated.

[0088] As an optional embodiment of the present disclosure, the specific method for obtaining the image to be evaluated and the generated keywords corresponding to the image to be evaluated has been described in S101 and will not be repeated here. This step can be implemented with reference to the step in S101, and the electronic device obtains the generated image to be evaluated and the generated keywords corresponding to the image to be evaluated.

[0089] S302 : For the image to be evaluated, using an evaluation algorithm corresponding to at least one mechanical evaluation index, to calculate a matching result between the image to be evaluated and the generated keyword.

[0090] As an optional embodiment of the present disclosure, an evaluation algorithm corresponding to at least one mechanical evaluation indicator is used, and a specific method for calculating the matching result between the image to be evaluated and the generated keyword has been described in S102 and will not be repeated here. This step can be implemented with reference to the step in S102, and one or a combination of several of the mechanical evaluation indicators listed in S102 is selected to obtain the matching result between the image to be evaluated and the generated keyword.

[0091] S303: input the image to be evaluated and the generated keywords into a trained visual question answering model, and evaluate the image to be evaluated from at least one preset evaluation dimension through the visual question answering model to obtain evaluation results corresponding to each evaluation dimension.

[0092] As an optional embodiment of the present disclosure, the simulated manual evaluation module is a trained visual question-answering model; the evaluating the image to be evaluated from at least one preset user-perceived evaluation dimension by the simulated manual evaluation module includes: asking questions to the image to be evaluated based on a plurality of preset manual evaluation-related questions by the visual question-answering model, and obtaining answers to the image to be evaluated by the visual question-answering model. Specifically, 9 manual evaluation-related questions of the visual question-answering model can be set as 9 detection points, represented by question, and questions are asked to the image to be evaluated, such as 1. "The image to be evaluated ”, 2. “What objects are there in the image to be evaluated?”, 3. “What actions appear in the image to be evaluated?”, 3. “What prepositions are there in the image to be evaluated?”, 4. “What adjectives are there in the image to be evaluated?”, 5. “How many different objects are there in the image to be evaluated?”, 6. “What is the area ratio of each object in the image to be evaluated?”, 7. “What objects have incorrect shapes in the image to be evaluated?”, 8. “What is the image style in the image to be evaluated?”, 9. “What is the aesthetic score of the image to be evaluated?” The objects include: people, plants, food, animals and other tangible objects. The visual question answering model answers the above 9 questions through machine learning, and at the same time obtains the above 9 objects from the biological objects. Extract reference answers corresponding to the multiple manual evaluation related questions from the generated keywords, compare the reference answers with the answers to the image to be evaluated to obtain multiple comparison results. For example, in question 4, the answer to the image to be evaluated is that there are three adjectives, and the reference answer in the generated keywords is 5 keywords. Then there are two error values ​​on question 4. In this way, 9 questions are compared to obtain multiple comparison results, and the multiple comparison results are mapped to at least one preset evaluation dimension to obtain the evaluation result corresponding to the at least one preset evaluation dimension. The multiple comparison results are classified according to the multiple preset evaluation dimensions; for each evaluation dimension, the various comparison results classified into the evaluation dimension are classified into the evaluation dimension. The comparison results are multiplied and summed with the corresponding weight coefficients to obtain the evaluation results corresponding to the evaluation dimensions. For example, the comparison results of questions 1, 2, 3, and 4 are weighted to obtain the evaluation results of the content correctness dimension, the comparison results of questions 5, 6, and 7 are weighted to obtain the evaluation results of the content relevance dimension, and the comparison results of questions 8 and 9 are weighted to obtain the evaluation results of the aesthetics dimension. Questions corresponding to any one of the three evaluation dimensions perceived by users can be selected to evaluate the image quality. Preferably, the three dimensions of content correctness, content relevance, and aesthetics are selected for quality evaluation.

[0093] S304, searching for a preset quality evaluation level corresponding to the matching result, and searching for a preset quality level corresponding to the evaluation result, and determining the preset quality level with the largest proportion as the quality evaluation result corresponding to the image to be evaluated.

[0094] As an optional embodiment of the present disclosure, the target quality evaluation corresponding to the matching result and the evaluation result can be searched in a preset quality score evaluation table. The specific quality score evaluation table has been described in step S104. The quality score evaluation table method described in step S104 is used for setting. According to the evaluation results of the content correctness dimension, content relevance dimension, and aesthetics dimension obtained in step S303, for example, the weighted sum of the content correctness dimension is 68 points, the weighted sum of the content relevance dimension is 80 points, and the weighted sum of the content aesthetics dimension is 60 points, then the corresponding target quality evaluation is searched according to the quality score evaluation table, and the target quality evaluation corresponding to the mechanical evaluation indicator and the content correctness, content relevance, and aesthetics is searched. The preset quality level with the largest proportion is determined as the quality evaluation result corresponding to the image to be evaluated, and the target quality evaluation is determined as the quality corresponding to the image to be evaluated.

[0095] In this embodiment, the scores of the three dimensions of content correctness, content relevance and aesthetics obtained by the visual question answering model are more in line with the user's perception to evaluate the image to be evaluated, compensate for the singleness of the mechanical evaluation indicators, and add rich multi-dimensional evaluation indicators. At the same time, the application of the visual question answering model frees up labor costs and avoids the uncertainty of the evaluation, making the evaluation results more reasonable and reliable.

[0096] like Figure 4 It is a schematic diagram of an application framework of an image quality assessment method provided by the first aspect of the embodiment of the present disclosure. Keywords are generated by user input, and AI draws a picture, namely, the image to be evaluated in this embodiment. The image to be evaluated is input for quality assessment, and an assessment algorithm corresponding to at least one mechanical assessment indicator is used to calculate the matching result between the image to be evaluated and the generated keyword, and the image is input into a manual assessment module at the same time. The manual assessment module provides two assessment methods in this embodiment. One is to directly calculate and analyze the indicators of the three dimensions of content correctness, content relevance, and aesthetics to obtain the assessment results of the assessment dimensions corresponding to user perception; the other is to input the image to be evaluated into a visual question-answering model, and the visual question-answering model evaluates the image to be evaluated from multiple preset evaluation dimensions to obtain the evaluation results corresponding to the content correctness, content relevance, and aesthetics, and find the target quality evaluation corresponding to the matching result and the evaluation result, and determine the target quality evaluation as the quality corresponding to the image to be evaluated, and the quality assessment ends.

[0097] Based on the same inventive concept, as an implementation of the above method, an embodiment of the present application also provides an image quality assessment device, which corresponds to the above method embodiment. For ease of reading, this embodiment will no longer repeat the details of the above method embodiment one by one, but it should be clear that the image quality assessment in this embodiment can correspond to all the contents in the above method embodiment.

[0098] The present application provides an image quality assessment device. Figure 5 is a schematic diagram of the structure of an image quality assessment device provided by the second aspect of the embodiment of the present disclosure, such as Figure 5 As shown, the information prompting device 500 includes:

[0099] An acquisition module 501 is used to acquire an image to be evaluated and a generated keyword corresponding to the image to be evaluated; the image to be evaluated is an image generated based on the generated keyword according to a preset text-generated image model;

[0100] A calculation module 502 is used to calculate a matching result between the image to be evaluated and the generated keyword by using an evaluation algorithm corresponding to at least one mechanical evaluation index for the image to be evaluated;

[0101] An evaluation module 503 is used to input the image to be evaluated and the generated keywords into a preset simulated manual evaluation module, and evaluate the image to be evaluated from at least one preset user-perceived evaluation dimension through the simulated manual evaluation module to obtain an evaluation result corresponding to the evaluation dimension;

[0102] The determination module 504 is used to find the preset quality evaluation level corresponding to the matching result and the preset quality level corresponding to the evaluation result, and determine the preset quality level with the largest proportion as the quality evaluation result corresponding to the image to be evaluated.

[0103] Based on the same inventive concept, an embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the computing device implements the image quality assessment method provided in the above embodiment.

[0104] Based on the same inventive concept, an embodiment of the present disclosure also provides an electronic device. Figure 6It is a structural schematic diagram of an electronic device provided by the fourth aspect of the embodiment of the present disclosure, and the electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workbenches, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0105] like Figure 6 As shown, the device 600 includes a computing unit 601, which can perform various appropriate actions and processes according to a computer program stored in a ROM (Read-Only Memory) 602 or a computer program loaded from a storage unit 608 to a RAM (Random Access Memory) 603. In the RAM 603, various programs and data required for the operation of the device 600 can also be stored. The computing unit 601, the ROM 602, and the RAM 603 are connected to each other via a bus 604. An I / O (Input / Output) interface 605 is also connected to the bus 604.

[0106] A number of components in the device 600 are connected to the I / O interface 605, including: an input unit 606, such as a keyboard, a mouse, etc.; an output unit 607, such as various types of displays, speakers, etc.; a storage unit 608, such as a disk, an optical disk, etc.; and a communication unit 609, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 609 allows the device 600 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0107] The computing unit 601 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 601 include, but are not limited to, a CPU (Central Processing Unit), a GPU (Graphic Processing Units), various dedicated AI (Artificial Intelligence) computing chips, various computing units running machine learning model algorithms, a DSP (Digital Signal Processor), and any appropriate processor, controller, microcontroller, etc. The computing unit 601 performs the various methods and processes described above, such as an image quality assessment method. For example, in some embodiments, the image quality assessment method may be implemented as a computer software program, which is tangibly included in a machine-readable medium, such as a storage unit 608. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 600 via the ROM 602 and / or the communication unit 609. When the computer program is loaded into the RAM 603 and executed by the computing unit 601, one or more steps of the method described above may be performed. Alternatively, in other embodiments, the computing unit 601 may be configured to execute the aforementioned communication method in any other appropriate manner (for example, by means of firmware).

[0108] Various embodiments of the systems and techniques described above herein may be implemented in digital electronic circuit systems, integrated circuit systems, FPGAs (Field Programmable Gate Arrays), ASICs (Application-Specific Integrated Circuits), ASSPs (Application Specific Standard Products), SOCs (System On Chips), CPLDs (Complex Programmable Logic Devices), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include: being implemented in one or more computer programs that may be executed and / or interpreted on a programmable system including at least one programmable processor that may be a dedicated or general-purpose programmable processor that may receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0109] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0110] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any suitable combination of the foregoing. More specific examples of machine-readable storage media may include electrical connections based on one or more lines, portable computer disks, hard disks, RAM, ROM, EPROM (Electrically Programmable Read-Only-Memory) or flash memory, optical fiber, CD-ROM (Compact Disc Read-Only Memory), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0111] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (Cathode-Ray Tube) or LCD (Liquid Crystal Display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0112] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: LAN (Local Area Network), WAN (Wide Area Network), the Internet, and blockchain networks.

[0113] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship between the client and the server is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services ("Virtual Private Server", or "VPS" for short). The server may also be a server of a distributed system, or a server combined with a blockchain.

[0114] It should be noted that artificial intelligence is a discipline that studies how computers can simulate certain human thought processes and intelligent behaviors (such as learning, reasoning, thinking, planning, etc.), and includes both hardware-level and software-level technologies. Artificial intelligence hardware technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, and big data processing; artificial intelligence software technologies mainly include computer vision technology, speech recognition technology, natural language processing technology, as well as machine learning / deep learning, big data processing technology, knowledge graph technology, and other major directions.

[0115] Based on the same inventive concept, an embodiment of the present application further provides a vehicle, which includes the image quality assessment device provided by the above embodiment or the electronic device provided by the above embodiment.

[0116] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0117] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A method for evaluating image quality, characterized in that: The method comprises: Acquire an image to be evaluated and a generated keyword corresponding to the image to be evaluated; the image to be evaluated is an image generated based on the generated keyword according to a preset text-generated image model; For the image to be evaluated, using an evaluation algorithm corresponding to at least one mechanical evaluation index to calculate a matching result between the image to be evaluated and the generated keyword; Inputting the image to be evaluated and the generated keywords into a preset simulated manual evaluation module, and evaluating the image to be evaluated from at least one preset user-perceived evaluation dimension through the simulated manual evaluation module to obtain an evaluation result corresponding to the evaluation dimension; The preset quality evaluation level corresponding to the matching result is searched, and the preset quality level corresponding to the evaluation result is searched, and the preset quality level with the largest proportion is determined as the quality evaluation result corresponding to the image to be evaluated.

2. The method according to claim 1, characterized in that The step of using an evaluation algorithm corresponding to at least one mechanical evaluation index to calculate a matching result between the image to be evaluated and the generated keyword includes: Performing feature vector conversion on the image to be evaluated and the generated keywords respectively to obtain an image feature vector corresponding to the image to be evaluated and a text feature vector corresponding to the generated keywords; The similarity between the image feature vector and the text feature vector is calculated, and the similarity is used as the matching result.

3. The method according to claim 1, characterized in that The evaluation dimensions include: at least one of content correctness, content relevance, and aesthetics; The step of evaluating the image to be evaluated from at least one preset user-perceived evaluation dimension by the simulated manual evaluation module includes: When the evaluation dimension includes the content correctness, searching whether there is an object corresponding to the generated keyword in the image to be evaluated, and determining the coverage of the generated keyword by the image to be evaluated; When the evaluation dimension includes the content relevance, comparing the generated keywords with the morphological parameters of each object in the image to be evaluated, and determining the number of objects with incorrect morphology in the image to be evaluated; When the evaluation dimension includes the aesthetics, it is determined whether the style of the image to be evaluated is consistent with the style in the generated keywords to obtain a corresponding style consistency result, and the aesthetics of the image to be evaluated is scored to obtain a corresponding aesthetics score.

4. The method according to claim 3, characterized in that When the evaluation dimension includes the content correctness, searching whether there is an object corresponding to the generated keyword in the image to be evaluated, and determining the coverage of the generated keyword by the image to be evaluated, includes: In view of the correctness of the content, performing an image-to-text operation on the image to be evaluated by using a preset image-to-text algorithm to obtain a text description corresponding to the image to be evaluated; The similarity between the text description and the generated keywords is calculated, and based on the similarity, the coverage of the image to be evaluated for the generated keywords is determined, and the coverage is determined as the evaluation result corresponding to the content correctness.

5. The method according to claim 3, characterized in that: When the evaluation dimension includes the content relevance, comparing the generated keywords with the morphological parameters of each object in the image to be evaluated to determine the number of objects with incorrect morphology in the image to be evaluated, comprises: According to the content relevance, a preset image recognition algorithm is used to perform object recognition on the image to be evaluated, so as to obtain morphological parameters of each object in the image to be evaluated; The preset morphological parameters of each object in the generated keyword are compared with the morphological parameters to determine the number of objects with incorrect morphology in the image to be evaluated, and the number of objects with incorrect morphology is determined as the evaluation result corresponding to the content relevance.

6. The method according to claim 3, characterized in that When the evaluation dimension includes the aesthetics, determining whether the style of the image to be evaluated is consistent with the style in the generated keywords to obtain a corresponding style consistency result, and scoring the aesthetics of the image to be evaluated to obtain a corresponding aesthetics score includes: With respect to the aesthetics, the style of the image to be evaluated is predicted by a style prediction algorithm, and it is determined whether the predicted style is consistent with the style in the generated keywords, thereby obtaining a corresponding style consistency result; Using a preset aesthetic scoring algorithm, the image to be evaluated is scored for aesthetics to obtain a corresponding aesthetic score; The style consistency result and the aesthetic score are determined as the evaluation result corresponding to the aesthetic.

7. The method according to claim 1, characterized in that The simulated manual evaluation module is a preset visual question-answering model; the step of evaluating the image to be evaluated from at least one preset user-perceived evaluation dimension by the simulated manual evaluation module includes: Performing image recognition on the image to be evaluated by using the visual question answering model to obtain answers to a plurality of preset manual evaluation related questions; Reference answers corresponding to the multiple manual evaluation related questions are extracted from the generated keywords, the reference answers are compared with the preset answers to the multiple manual evaluation related questions to obtain multiple comparison results, the multiple comparison results are mapped to at least one preset evaluation dimension, and the evaluation results corresponding to the at least one preset evaluation dimension are obtained.

8. The method according to claim 7, characterized in that Mapping the multiple comparison results to at least one preset evaluation dimension to obtain an evaluation result corresponding to the at least one preset evaluation dimension includes: Classify multiple comparison results according to multiple preset evaluation dimensions; For each evaluation dimension, each comparison result classified into the evaluation dimension is multiplied and summed with the corresponding weight coefficient to obtain the evaluation result corresponding to the evaluation dimension.

9. An image quality assessment device, characterized in that: include: An acquisition module, used to acquire an image to be evaluated and a generated keyword corresponding to the image to be evaluated; The image to be evaluated is an image generated based on the generated keywords according to a preset text-generated image model; A calculation module, configured to calculate a matching result between the image to be evaluated and the generated keyword by using an evaluation algorithm corresponding to at least one mechanical evaluation index for the image to be evaluated; An evaluation module, configured to input the image to be evaluated and the generated keywords into a preset simulated manual evaluation module, and evaluate the image to be evaluated from at least one preset user-perceived evaluation dimension through the simulated manual evaluation module to obtain an evaluation result corresponding to the evaluation dimension; The determination module is used to find the preset quality evaluation level corresponding to the matching result and the preset quality level corresponding to the evaluation result, and determine the preset quality level with the largest proportion as the quality evaluation result corresponding to the image to be evaluated.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method described in any one of claims 1 to 8 are implemented.

11. An electronic device, characterized in that: include: a memory having a computer program stored thereon; A processor, configured to execute the computer program in the memory to implement the steps of the method according to any one of claims 1 to 8.

12. A vehicle, characterized in that: The device comprises the image quality assessment device as claimed in claim 9.

Images