人工智能生成图像质量评估综述：从评估维度到模型方法

doi:10.11996/JG.j.2095-302X.2026050941

图学学报 ›› 2026, Vol. 47 ›› Issue (5): 941-953.DOI: 10.11996/JG.j.2095-302X.2026050941

人工智能生成图像质量评估综述：从评估维度到模型方法

杨立然¹^,², 李可心¹, 张诗吟³, 苏攀¹^,⁴()

¹ 华北电力大学计算机系，河北保定 071003
² 复杂能源系统智能计算教育部工程研究中心，河北保定 071003
³ 华北电力大学自动化系，河北保定 071003
⁴ 河北省能源电力知识计算重点实验室，河北保定 071003

收稿日期:2026-03-30 接受日期:2026-07-03 出版日期:2026-10-31 发布日期:2026-10-26
通讯作者:苏攀，E-mail：supan@ncepu.edu.cn
基金资助:
河北省自然科学基金(F2024502002);国家自然科学基金青年科学基金(62403199);中央高校基本科研业务费专项资金(2026MS186)

A review of artificial intelligence-generated image quality assessment: from evaluation dimensions to model methods

YANG Liran¹^,², LI Kexin¹, ZHANG Shiyin³, SU Pan¹^,⁴()

¹ Department of Computer, North China Electric Power University, Baoding Hebei 071003, China
² Engineering Research Center of Intelligent Computing for Complex Energy Systems, Ministry of Education, Baoding Hebei 071003, China
³ Department of Automation, North China Electric Power University, Baoding Hebei 071003, China
⁴ Hebei Key Laboratory of Knowledge Computing for Energy and Power, Baoding Hebei 071003, China

Received:2026-03-30 Accepted:2026-07-03 Published:2026-10-31 Online:2026-10-26
Contact: SU Pan，E-mail：supan@ncepu.edu.cn
Supported by:
Hebei Natural Science Foundation(F2024502002);National Natural Science Foundation of China(62403199);Fundamental Research Funds for the Central Universities(2026MS186)

摘要/Abstract

摘要：

图像质量评估(IQA)是图像处理与计算机视觉领域中极具研究价值的重要课题，目标是建立以人类视觉习惯为基准的图像质量评价模型。现在人工智能生成内容(AIGC)发展迅速，在视觉生成上取得明显进展，尤其在图像生成方面表现突出。但受到图像生成的固有限制，人工智能生成图像(AIGI)普遍存在语义偏移、结构不合理和视觉伪影等质量问题，传统IQA方法不能充分描述AIGI的特有缺陷，也无法很好地满足其评价需求。AIGI质量评估(AIGIQA)遂成为计算机视觉领域的新兴焦点。为此，围绕AIGIQA现有技术体系与研究方法开展全面梳理、系统调研与综述分析。首先，从主流图像生成架构出发，梳理生成对抗网络、自回归生成模型和扩散模型3类生成模型，并分析各类模型导致的图像质量缺陷；然后，对AIGI的质量评估维度进行归纳总结，从技术问题、AI伪影、不自然性、差异性和美学5个维度阐述评价内涵；接着对AIGIQA常用模型性能评估指标做归纳梳理，并对当前公开数据集进行了介绍和对比分析；此外，回顾了传统IQA方法的发展状况及其评估AIGI时存在的局限性；对现有AIGIQA方法进行分类，重点总结了基于跨模态对齐和基于多模态大语言模型(MLLM)二类代表性方法的技术路线、核心思想和局限性。最后说明了AIGIQA在质量表征、数据支撑、模型构建及可解释性等方面面临的机遇与挑战，指出了未来潜在探索方向，为该领域的后续研究提供了参考性建议。

关键词: 人工智能生成图像, 图像质量评估, 评估维度, 评估指标, 评估方法

Abstract:

Image Quality Assessment (IQA) is a valuable and vital research topic in image processing and computer vision, aiming to construct image quality assessment models in accordance with human visual perception. With the rapid advancement of Artificial Intelligence-Generated Content (AIGC), remarkable progress has been made in visual generation, particularly in image synthesis. However, due to the inherent limitations of image generation mechanisms, Artificial Intelligence-Generated Images (AIGI) often exhibit various quality defects, including semantic drift, unreasonable structures, and visual artifacts. Traditional IQA methods fail to adequately capture these unique defects, thus falling short in meeting the evaluation needs for AIGI. Consequently, AIGI Quality Assessment (AIGIQA) has emerged as a prominent research topic in computer vision. Existing technical frameworks and research methodologies of AIGIQA were comprehensively reviewed, systematically investigated, and thoroughly analyzed. First, three representative image generation paradigms, namely generative adversarial networks, autoregressive models, and diffusion models, were reviewed, and the typical image quality defects associated with different generation mechanisms were analyzed. Next, the quality evaluation dimensions of AIGI were summarized from five perspectives, including technical issues, AI artifacts, unnaturalness, discrepancy, and aesthetics. Furthermore, commonly used performance evaluation metrics for AIGIQA models were reviewed, and publicly available datasets were introduced and compared. In addition, the development of traditional IQA methods was revisited, and their limitations in assessing AIGI were thoroughly discussed. Existing AIGIQA approaches were further categorized, with particular emphasis on the technical frameworks, core ideas, and limitations of representative methods based on cross-modal alignment and Multimodal Large Language Models (MLLMs). Finally, the opportunities and challenges faced by AIGIQA in quality representation, dataset support, model construction, and interpretability were discussed, and potential directions for future research were identified, providing valuable references for subsequent studies in this field.

Key words: artificial intelligence-generated images, image quality assessment, evaluation dimensions, evaluation metrics, assessment methods

中图分类号:

杨立然, 李可心, 张诗吟, 苏攀. 人工智能生成图像质量评估综述：从评估维度到模型方法[J]. 图学学报, 2026, 47(5): 941-953.

YANG Liran, LI Kexin, ZHANG Shiyin, SU Pan. A review of artificial intelligence-generated image quality assessment: from evaluation dimensions to model methods[J]. Journal of Graphics, 2026, 47(5): 941-953.

图/表 9

参考文献 102

[1]	LI C Y, ZHANG Z C, WU H N, et al. AGIQA-3K: an open database for AI-generated image quality assessment[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2024, 34(8): 6833-6846. DOI URL
[2]	邹良涛, 蒋刚毅, 郁梅, 等. 基于张量域感知特征的无参考高动态范围图像质量评价[J]. 计算机辅助设计与图形学学报, 2018, 30(10): 1850-1858.
	ZOU L T, JIANG G Y, YU M, et al. No-reference image quality assessment of high dynamic range image based on tensor domain perceptual features[J]. Journal of Computer-Aided Design & Computer Graphics, 2018, 30(10): 1850-1858 (in Chinese).
[3]	GOODFELLOW I J, POUGET-ABADIE J, MIRZA M, et al. Generative adversarial nets[C]// The 28th International Conference on Neural Information Processing Systems. Cambridge: MIT Press, 2014: 2672-2680.
[4]	LAROCHELLE H, MURRAY I. The neural autoregressive distribution estimator[EB/OL]. [2025-11-30]. https://dblp.org/db/journals/jmlr/jmlrp15.html#journals/jmlr/LarochelleM11.
[5]	HO J, JAIN A, ABBEEL P. Denoising diffusion probabilistic models[C]// The 34th International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2020: 574.
[6]	杨皓然, 马瑞强, 王钢, 等. 人工智能生成图像检测综述[J]. 计算机技术与发展, 2025, 35(11): 1-11.
	YANG H R, MA R Q, WANG G, et al. Review of AI-generated image detection[J]. Computer Technology and Development, 2025, 35(11): 1-11 (in Chinese).
[7]	XIAO J, BI X J. Model-guided generative adversarial networks for unsupervised fine-grained image generation[J]. IEEE Transactions on Multimedia, 2024, 26: 1188-1199. DOI URL
[8]	TANG H, LIU H, XU D, et al. AttentionGAN: unpaired image-to-image translation using attention-guided generative adversarial networks[J]. IEEE Transactions on Neural Networks and Learning Systems, 2023, 34(4): 1972-1987. DOI URL
[9]	PAN X G, TEWARI A, LEIMKÜHLER T, et al. Drag your GAN: interactive point-based manipulation on the generative image manifold[C]// ACM SIGGRAPH 2023 Conference Proceedings. New York: Association for Computing Machinery, 2023: 78.
[10]	向婷, 唐卓, 郑佳丽, 等. 基于生成模型的图像数据增强方法综述[J]. 图学学报, 2026, 47(2): 235-250. DOI
	XIANG T, TANG Z, ZHENG J L, et al. A review of image data augmentation based on generative models[J]. Journal of Graphics, 2026, 47(2): 235-250 (in Chinese). DOI
[11]	RADFORD A, METZ L, CHINTALA S. Unsupervised representation learning with deep convolutional generative adversarial networks[EB/OL]. [2025-11-14]. https://arxiv.org/abs/1511.06434.
[12]	KARRAS T, AILA T, LAINE S, et al. Progressive growing of GANs for improved quality, stability, and variation[EB/OL]. [2025-11-14]. https://arxiv.org/abs/1710.10196.
[13]	KARRAS T, LAINE S, AILA T. A style-based generator architecture for generative adversarial networks[C]// 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2019: 4396-4405.
[14]	VAN DEN OORD A, KALCHBRENNER N, KAVUKCUOGLU K. Pixel recurrent neural networks[EB/OL]. [2025-11-30]. https://dl.acm.org/doi/abs/10.5555/3045390.3045575.
[15]	HOCHREITER S, SCHMIDHUBER J. Long short-term memory[J]. Neural Computation, 1997, 9(8): 1735-1780. DOI PMID
[16]	VAN DEN OORD A, KALCHBRENNER N, ESPEHOLT L, et al. Conditional image generation with PixelCNN decoders[C]// The 30th International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2016: 4797-4805.
[17]	VASWANI A, SHAZEER N, PARMAR N, et al. Attention is all you need[C]// The 31st International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2017: 6000-6010.
[18]	DING M, YANG Z Y, HONG W Y, et al. CogView: mastering text-to-image generation via transformers[EB/OL]. [2025-11-30]. https://arxiv.org/abs/2105.13290.
[19]	ROMBACH R, BLATTMANN A, LORENZ D, et al. High-resolution image synthesis with latent diffusion models[C]// 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2022: 10674-10685.
[20]	PODELL D, ENGLISH Z, LACEY K, et al. SDXL: improving latent diffusion models for high-resolution image synthesis[EB/OL]. [2025-11-30]. https://arxiv.org/abs/2307.01952.
[21]	PEEBLES W, XIE S N. Scalable diffusion models with transformers[C]// 2023 IEEE/CVF International Conference on Computer Vision. New York: IEEE Press, 2023: 4195-4205.
[22]	RAMESH A, DHARIWAL P, NICHOL A, et al. Hierarchical text-conditional image generation with CLIP latents[EB/OL]. [2025-06-06]. https://arxiv.org/abs/2204.06125.
[23]	NICHOL A Q, DHARIWAL P, RAMESH A, et al. GLIDE: towards photorealistic image generation and editing with text-guided diffusion models[C]// The 39th International Conference on Machine Learning. New York: PMLR, 2022, 162: 16784-16804.
[24]	SAHARIA C, CHAN W, SAXENA S, et al. Photorealistic text-to-image diffusion models with deep language understanding[C]// The 36th International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2022: 2643.
[25]	叶文龙, 陈斌. PanoLoRA:基于Stable Diffusion的全景图像生成的高效微调方法[J]. 图学学报, 2025, 46(5): 980-989. DOI
	YE W L, CHEN B. PanoLoRA: an efficient finetuning method for panoramic image generation based on Stable Diffusion[J]. Journal of Graphics, 2025, 46(5): 980-989 (in Chinese). DOI
[26]	黄英吉, 钟猛猛, 张庭霨, 等. 基于客观评价的图像质量评估方法综述[J]. 电脑知识与技术, 2021, 17(28): 92-94.
	HUANG Y J, ZHONG M M, ZHANG T W, et al. Overview of image quality evaluation methods based on objective evaluation[J]. Computer Knowledge and Technology, 2021, 17(28): 92-94 (in Chinese).
[27]	邢润媚, 常升龙, 何宽, 等. AIGC图像质量评估指标研究[J]. 南京信息工程大学学报, 2025, 17(1): 63-73.
	XING R M, CHANG S L, HE K, et al. AIGC image quality evaluation indicators[J]. Journal of Nanjing University of Information Science & Technology, 2025, 17(1): 63-73 (in Chinese).
[28]	ZHANG Z C, LI C Y, SUN W, et al. A perceptual quality assessment exploration for AIGC images[C]// 2023 IEEE International Conference on Multimedia and Expo Workshops. New York: IEEE Press, 2023: 440-445.
[29]	WANG J R, DUAN H Y, LIU J, et al. AIGCIQA2023: a large-scale image quality assessment database for AI generated images: from the perspectives of quality, authenticity and correspondence[C]// The 3rd CAAI International Conference on Artificial Intelligence. Cham: Springer, 2024: 46-57.
[30]	CHEN Z J, SUN W, WU H N, et al. Study of subjective and objective naturalness assessment of AI-generated images[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2025, 35(4): 3573-3588. DOI URL
[31]	高丽霞, 刘国军, 张选德. 快速提升的结构相似度IQA算法[J]. 计算机辅助设计与图形学学报, 2017, 29(7): 1314-1323.
	GAO L X, LIU G J, ZHANG X D. Fast lifting IQA algorithm of the structure similarity index[J]. Journal of Computer-Aided Design & Computer Graphics, 2017, 29(7): 1314-1323 (in Chinese).
[32]	王志明. 无参考图像质量评价综述[J]. 自动化学报, 2015, 41(6): 1062-1079.
	WANG Z M. Review of No-reference image quality assessment[J]. Acta Automatica Sinica, 2015, 41(6): 1062-1079 (in Chinese).
[33]	KIRSTAIN Y, POLYAK A, SINGER U, et al. Pick-a-pic: an open dataset of user preferences for text-to-image generation[C]// The 37th International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2023: 1594.
[34]	LI C Y, KOU T C, GAO Y X, et al. AIGIQA-20K: a large database for AI-generated image quality assessment[C]// 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops. New York: IEEE Press, 2024: 6327-6336.
[35]	YUAN J Q, LI J H, YANG F Y, et al. PKU-AIGIQA-4K: a perceptual quality assessment database for both text-to-image and image-to-image AI-generated images[C]// 2025 IEEE/CVF International Conference on Computer Vision Workshops. New York: IEEE Press, 2025: 3362-3371.
[36]	XU T, ZHANG P C, HUANG Q Y, et al. AttnGAN: fine-grained text to image generation with attentional generative adversarial networks[C]// 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2018: 1316-1324.
[37]	ZHOU Y F, ZHANG R Y, CHEN C Y, et al. Towards language-free training for text-to-image generation[C]// 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2022: 17886-17896.
[38]	BAO F, NIE S, XUE K W, et al. One transformer fits all distributions in multi-modal diffusion at scale[EB/OL]. [2025-11-30]. https://dl.acm.org/doi/10.5555/3618408.3618480.
[39]	ZHANG L, RAO A, AGRAWALA M. Adding conditional control to text-to-image diffusion models[C]// 2023 IEEE/CVF International Conference on Computer Vision. New York: IEEE Press, 2023: 3836-3847.
[40]	CHEN J S, YU J C, GE C J, et al. Pixart-α: fast training of diffusion transformer for photorealistic text-to-image synthesis[EB/OL]. [2025-11-14]. https://arxiv.org/abs/2310.00426.
[41]	GUPTA Y, JADDIPAL V V, PRABHALA H, et al. Progressive knowledge distillation of stable diffusion XL using layer level loss[EB/OL]. 2024 [2025-11-14]. https://arxiv.org/abs/2401.02677.
[42]	LUO S M, TAN Y Q, PATIL S, et al. LCM-LoRA: a universal stable-diffusion acceleration module[EB/OL]. [2025-11-14].https://arxiv.org/abs/2311.05556.
[43]	SAUER A, LORENZ D, BLATTMANN A, et al. Adversarial diffusion distillation[C]// The 18th European Conference on Computer Vision. Cham: Springer, 2024: 87-103.
[44]	BETKER J, GOH G, JING L, et al. Improving image generation with better captions[EB/OL]. [2025-12-14]. https://api.semanticscholar.org/CorpusID:264403242.
[45]	ZHAN F N, YU Y C, WU R L, et al. Bi-level feature alignment for versatile image translation and manipulation[C]// The 17th European Conference on Computer Vision. Cham: Springer, 2022: 224-241.
[46]	KWON G, YE J C. Diffusion-based image translation using disentangled style and content representation[EB/OL]. [2025-12-14]. https://arxiv.org/abs/2209.15264.
[47]	PATASHNIK O, WU Z Z, SHECHTMAN E, et al. StyleCLIP: text-driven manipulation of StyleGAN imagery[C]// 2021 IEEE/CVF International Conference on Computer Vision. New York: IEEE Press, 2021: 2065-2074.
[48]	JIANG C, GAO F, MA B, et al. Masked and adaptive transformer for exemplar based image translation[C]// 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2023: 22418-22427.
[49]	ZHANG P, ZHANG B, CHEN D, et al. Cross-domain correspondence learning for exemplar-based image translation[C]// 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2020: 5142-5152.
[50]	LUGMAYR A, DANELLJAN M, ROMERO A, et al. RePaint: inpainting using denoising diffusion probabilistic models[C]// 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2022: 11451-11461.
[51]	LI W B, LIN Z, ZHOU K, et al. MAT: mask-aware transformer for large hole image inpainting[C]// 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2022: 10748-10758.
[52]	WANG H Z, ZHAI D M, LIU X M, et al. Unsupervised deep exemplar colorization via pyramid dual non-local attention[J]. IEEE Transactions on Image Processing, 2023, 32: 4114-4127. DOI PMID
[53]	KANG X Y, YANG T, OUYANG W Q, et al. DDColor: towards photo-realistic image colorization via dual decoders[C]// 2023 IEEE/CVF International Conference on Computer Vision. New York: IEEE Press, 2023: 328-338.
[54]	XIA M H, HU W B, WONG T T, et al. Disentangled image colorization via global anchors[J]. ACM Transactions on Graphics, 2022, 41(6): 204.
[55]	BROOKS T, HOLYNSKI A, EFROS A A. InstructPix2Pix: learning to follow image editing instructions[C]// 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2023: 18392-18402.
[56]	ZHANG K, MO L B, CHEN W H, et al. MagicBrush: a manually annotated dataset for instruction-guided image editing[C]// The 37th International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2023: 1365.
[57]	陈健, 李诗云, 林丽, 等. 模糊失真图像无参考质量评价综述[J]. 自动化学报, 2022, 48(3): 689-711.
	CHEN J, LI S Y, LIN L, et al. A review on no-reference quality assessment for blurred image[J]. Acta Automatica Sinica, 2022, 48(3): 689-711 (in Chinese).
[58]	WANG Z, BOVIK A C, SHEIKH H R, et al. Image quality assessment: from error visibility to structural similarity[J]. IEEE Transactions on Image Processing, 2004, 13(4): 600-612. DOI PMID
[59]	金剑秋, 周天成, 胡紫豪, 等. SAM2ICAA:基于视觉大模型的无参考图像色彩准确性评估方法[EB/OL]. [2026-01-30]. https://link.cnki.net/urlid/11.2925.tp.20260212.1707.005.
	JIN J Q, ZHOU T C, HU Z H, et al. SAM2ICAA:no-reference image color accuracy assessment via vision foundation models[EB/OL]. [2026-01-30]. https://link.cnki.net/urlid/11.2925.tp.20260212.1707.005(in Chinese).
[60]	MITTAL A, SOUNDARARAJAN R, BOVIK A C. Making a “completely blind” image quality analyzer[J]. IEEE Signal Processing Letters, 2013, 20(3): 209-212. DOI URL
[61]	KANG L, YE P, LI Y, et al. Convolutional neural networks for no-reference image quality assessment[C]// 2024 IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2014: 1733-1740.
[62]	LIU X L, VAN DE WEIJER J, BAGDANOV A D. RankIQA: learning from rankings for no-reference image quality assessment[C]// 2017 IEEE International Conference on Computer Vision. New York: IEEE Press, 2017: 1040-1049.
[63]	ZHANG W X, MA K D, YAN J, et al. Blind image quality assessment using a deep bilinear convolutional neural network[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2020, 30(1): 36-47. DOI URL
[64]	SU S L, YAN Q S, ZHU Y, et al. Blindly assess image quality in the wild guided by a self-adaptive hyper network[C]// 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2020: 3664-3673.
[65]	HE K M, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition[C]// 2016 IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2016: 770-778.
[66]	SUN W, DUAN H Y, MIN X K, et al. Blind quality assessment for in-the-wild images via hierarchical feature fusion strategy[C]// 2022 IEEE International Symposium on Broadband Multimedia Systems and Broadcasting. New York: IEEE Press, 2022: 1-6.
[67]	RADFORD A, KIM J W, HALLACY C, et al. Learning transferable visual models from natural language supervision[EB/OL]. [2025-11-30]. https://dblp.org/db/conf/icml/icml2021.html#conf/icml/RadfordKHRGASAM21.
[68]	LI J N, LI D X, XIONG C M, et al. BLIP: bootstrapping language-image pre-training for unified vision-language understanding and generation[EB/OL]. [2025-11-30]. https://dblp.org/db/conf/icml/icml2022.html#conf/icml/0001LXH22.
[69]	DOSOVITSKIY A, BEYER L, KOLESNIKOV A, et al. An image is worth 16x16 words:transformers for image recognition at scale[EB/OL]. [2025-12-14]. https://arxiv.org/html/2010.11929v2.
[70]	ZHANG W X, ZHAI G T, WEI Y, et al. Blind image quality assessment via vision-language correspondence: a multitask learning perspective[C]// 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2023: 14071-14081.
[71]	SHI J S, GAO P, SMOLIC A. Blind image quality assessment via transformer predicted error map and perceptual quality token[J]. IEEE Transactions on Multimedia, 2023, 26: 4641-4651. DOI URL
[72]	NARVEKAR N D, KARAM L J. A no-reference perceptual image sharpness metric based on a cumulative probability of blur detection[C]// 2009 International Workshop on Quality of Multimedia Experience. New York: IEEE Press, 2009: 87-91.
[73]	GU K, ZHAI G T, YANG X K, et al. Hybrid no-reference quality metric for singly and multiply distorted images[J]. IEEE Transactions on Broadcasting, 2014, 60(3): 555-567. DOI URL
[74]	YAN J, LI J, FU X. No-reference quality assessment of contrast-distorted images using contrast enhancement[EB/OL]. [2025-12-14]. https://arxiv.org/abs/1904.08879.
[75]	SALIMANS T, GOODFELLOW I, ZAREMBA W, et al. Improved techniques for training GANs[C]// The 30th International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2016: 2234-2242.
[76]	HEUSEL M, RAMSAUER H, UNTERTHINER T, et al. GANs trained by a two time-scale update rule converge to a local nash equilibrium[C]// The 31st International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2017: 6629-6640.
[77]	BIŃKOWSKI M, SUTHERLAND D J, ARBEL M, et al. Demystifying MMD GANs[EB/OL]. [2025-12-14]. https://arxiv.org/abs/1801.01401.
[78]	XUE W F, MOU X Q, ZHANG L, et al. Blind image quality assessment using joint statistics of gradient magnitude and laplacian features[J]. IEEE Transactions on Image Processing, 2014, 23(11): 4850-4862. DOI PMID
[79]	MIN X K, ZHAI G T, GU K, et al. Blind image quality estimation via distortion aggravation[J]. IEEE Transactions on Broadcasting, 2018, 64(2): 508-517. DOI URL
[80]	WANG J Y, CHAN K C K, LOY C C. Exploring CLIP for assessing the look and feel of images[C]// The AAAI Conference on Artificial Intelligence. Washington: AAAI Press, 2023, 37(2): 2555-2563.
[81]	HESSEL J, HOLTZMAN A, FORBES M, et al. CLIPScore: a reference-free evaluation metric for image captioning[EB/OL]. [2025-11-30]. https://aclanthology.org/2021.emnlp-main.595/.
[82]	XU J Z, LIU X, WU Y C, et al. ImageReward: learning and evaluating human preferences for text-to-image generation[C]// The 37th International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2023: 700.
[83]	WU X S, SUN K Q, ZHU F, et al. Human preference score: better aligning text-to-image models with human preference[C]// 2023 IEEE/CVF International Conference on Computer Vision. New York: IEEE Press, 2023: 2096-2105.
[84]	TANG Z C, WANG Z C, PENG B, et al. CLIP-AGIQA: boosting the performance of AI-generated image quality assessment with CLIP[C]// The 27th International Conference on Pattern Recognition. Cham: Springer, 2024: 48-61.
[85]	PENG F, FU H Y, MING A L, et al. AIGC image quality assessment via image-prompt correspondence[C]// 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops. New York: IEEE Press, 2024: 6432-6441.
[86]	XIA J L, HE L H, DENG C, et al. AI-generated image quality assessment based on task-specific prompt and multi-granularity similarity[J]. IEEE Transactions on Image Processing, 2025, 34: 8216-8228. DOI URL
[87]	ZHOU T W, TAN S B, ZHOU W, et al. Adaptive mixed-scale feature fusion network for blind AI-generated image quality assessment[J]. IEEE Transactions on Broadcasting, 2024, 70(3): 833-843. DOI URL
[88]	YU Z H, GUAN F B, LU Y T, et al. SF-IQA: quality and similarity integration for AI generated image quality assessment[C]// 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops. New York: IEEE Press, 2024: 6692-6701.
[89]	YANG J F, FU J, ZHANG W, et al. MoE-AGIQA: mixture- of-experts boosted visual perception-driven and semantic- aware quality assessment for AI-generated images[C]// 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops. New York: IEEE Press, 2024: 6395-6404.
[90]	杨浩中, 孔笑宇, 辜睿坤, 等. 虚拟现实中的大模型技术研究进展与趋势[J]. 图学学报, 2024, 45(6): 1117-1131. DOI
	YANG H Z, KONG X Y, GU R K, et al. Research progress and trends in large model technologies for virtual reality[J]. Journal of Graphics, 2024, 45(6): 1117-1131 (in Chinese). DOI
[91]	WANG P Y, SUN W, ZHANG Z C, et al. Large multi-modality model assisted AI-generated image quality assessment[C]// The 32nd ACM International Conference on Multimedia. New York: ACM, 2024: 7803-7812.
[92]	WANG J R, DUAN H Y, ZHAI G T, et al. Quality assessment for AI generated images with instruction tuning[J]. IEEE Transactions on Multimedia, 2026: 1-16.
[93]	CUI C, CHEN K J, WEI Z H, et al. M3-AGIQA:multimodal, multi-round, multi-aspect AI-generated image quality assessment[EB/OL]. [2025-12-14]. https://arxiv.org/abs/2502.15167.
[94]	LI Q, YAN Q S, HUANG H J, et al. Text-visual semantic constrained AI-generated image quality assessment[C]// The 33rd ACM International Conference on Multimedia. New York: ACM, 2025: 6958-6966.
[95]	QU B W, LI H H, GAO W. Bringing textual prompt to AI-generated image quality assessment[C]// 2024 IEEE International Conference on Multimedia and Expo. New York: IEEE Press, 2024: 1-6.
[96]	FU J, ZHOU W, JIANG Q P, et al. Vision-language consistency guided multi-modal prompt learning for blind AI generated image quality assessment[J]. IEEE Signal Processing Letters, 2024, 31: 1820-1824. DOI URL
[97]	CHEN Z W, QIN H N, WANG J, et al. PromptIQA: boosting the performance and generalization for No-reference image quality assessment via prompts[C]// The 18th European Conference on Computer Vision. Cham: Springer, 2024: 247-264.
[98]	GAO M J, ZHANG Q R, SONG C Y, et al. Hierarchical prompt engineering and task-differentiated low-rank adaptation for artificial intelligence-generated content image quality assessment[J]. Information, 2025, 16(11): 1006. DOI URL
[99]	ZHOU T W, TAN S B, LI L D, et al. Cross-modality interactive attention network for AI-generated image quality assessment[J]. Pattern Recognition, 2025, 167: 111693. DOI URL
[100]	PAN Z Q, YANG Y, YUAN F, et al. Assessing AI-generated image quality using a cross-modal hierarchical perception network[J]. IEEE Transactions on Broadcasting, 2026, 72(1): 291-302. DOI URL
[101]	LU Y, HAN S Y, ZHOU Z L, et al. A deep cross-modal prompt learning network for artificial intelligence generated image quality assessment[J]. Displays, 2026, 91: 103208. DOI URL
[102]	LI Y H, WU S J, SUN W, et al. AGHI-QA: a subjective-aligned dataset and metric for AI-generated human images[EB/OL]. [2025-12-14]. https://arxiv.org/abs/2504.21308.

评估维度	数据集标注	说明
技术问题(模糊、噪声)	感知质量/技术质量	表征图像低层视觉信号的退化，主要体现在图像清晰度与细节表达，与感知质量和技术质量评分相关
AI伪影(重复纹理、畸变)	感知质量/真实性/合理性/自然性	由生成机制产生的结构异常与重复模式，同时体现在视觉质量与自然属性相关评价维度
不自然性(光照/结构异常)	真实性/合理性/自然性/感知质量	反映图像对真实世界物理规律与结构关系的偏离，主要体现在真实性与自然性评价中
差异性(语义偏移)	图文一致性	表征图像内容与文本语义之间的不一致性，与图文一致性评价结果相关
美学(构图、风格)	感知质量	体现图像在构图与视觉表现层面的高层感知属性，综合反映在整体感知质量评分中

评估维度	数据集标注	说明
技术问题(模糊、噪声)	感知质量/技术质量	表征图像低层视觉信号的退化，主要体现在图像清晰度与细节表达，与感知质量和技术质量评分相关
AI伪影(重复纹理、畸变)	感知质量/真实性/合理性/自然性	由生成机制产生的结构异常与重复模式，同时体现在视觉质量与自然属性相关评价维度
不自然性(光照/结构异常)	真实性/合理性/自然性/感知质量	反映图像对真实世界物理规律与结构关系的偏离，主要体现在真实性与自然性评价中
差异性(语义偏移)	图文一致性	表征图像内容与文本语义之间的不一致性，与图文一致性评价结果相关
美学(构图、风格)	感知质量	体现图像在构图与视觉表现层面的高层感知属性，综合反映在整体感知质量评分中

数据集	图像数量	生成模型数量	评分类型	评分区间	评分标注维度	说明
AGIQA-1K^[28]	1 080	2	MOS	[0,5]	质量	最早提出，规模较小且标注有限
AGIQA-3K^[1]	2 982	6	MOS	[0,5]	质量、图文对齐	引入多类生成模型，早期生成模型性能不及前沿水平
AIGCIQA2023^[29]	2 400	6	MOS	[0,100]	质量、真实性、图文对齐	提供三维主观评分，数据来源与生成条件相对集中
Pick-A-Pic^[33]	500 000	3	成对偏好(Pairwise preference)	—	人类主观偏好	大规模人类偏好数据集，难以直接用于回归任务
AIGIQA-20K^[34]	20 000	15	MOS	[0,5]	质量和图文对齐综合考量	大规模的细粒度数据库，多维因素被压缩为单一分数
PKU-AIGIQA-4K^[35]	4 000	3	MOS	[0,5]	质量、真实性、图文对齐	支持图像到图像和文本到图像场景，生成模型覆盖范围窄
AGIN^[30]	6 049	18	MOS	[0,5]	技术质量、合理性、自然性	第一个研究图像自然性问题的数据集，侧重特定任务场景

数据集	图像数量	生成模型数量	评分类型	评分区间	评分标注维度	说明
AGIQA-1K^[28]	1 080	2	MOS	[0,5]	质量	最早提出，规模较小且标注有限
AGIQA-3K^[1]	2 982	6	MOS	[0,5]	质量、图文对齐	引入多类生成模型，早期生成模型性能不及前沿水平
AIGCIQA2023^[29]	2 400	6	MOS	[0,100]	质量、真实性、图文对齐	提供三维主观评分，数据来源与生成条件相对集中
Pick-A-Pic^[33]	500 000	3	成对偏好(Pairwise preference)	—	人类主观偏好	大规模人类偏好数据集，难以直接用于回归任务
AIGIQA-20K^[34]	20 000	15	MOS	[0,5]	质量和图文对齐综合考量	大规模的细粒度数据库，多维因素被压缩为单一分数
PKU-AIGIQA-4K^[35]	4 000	3	MOS	[0,5]	质量、真实性、图文对齐	支持图像到图像和文本到图像场景，生成模型覆盖范围窄
AGIN^[30]	6 049	18	MOS	[0,5]	技术质量、合理性、自然性	第一个研究图像自然性问题的数据集，侧重特定任务场景

类型	方法	AGIQA-3K
类型	方法	SRCC	KRCC	PLCC
基于手工制作	DSIQA^[72](2009)	0.495 5	0.340 3	0.548 8
	NIQE^[60](2012)	0.562 3	0.387 6	0.517 1
	SISBLIM^[73](2014)	0.547 9	0.378 8	0.647 7
	CEIQ^[74](2019)	0.322 8	0.222 0	0.416 6
基于分布距离	ICS^[75](2016)	0.093 1	0.062 6	0.096 4
	FID^[76](2017)	0.173 3	0.115 8	0.186 0
	KID^[77](2018)	0.102 3	0.069 2	0.078 6
基于SVR	GMLF^[78](2014)	0.698 7	0.511 9	0.818 1
基于SVR	BMPRI^[79](2018)	0.679 4	0.497 6	0.791 2
基于深度学习方法	CNNIQA^[61](2014)	0.747 8	0.558 0	0.846 9
	DBCNN^[63](2018)	0.820 7	0.633 6	0.875 9
	HyperNet^[64](2020)	0.835 5	0.648 8	0.890 3
	CLIPIQA^[80](2022)	0.842 6	0.646 8	0.805 3

人工智能生成图像质量评估综述：从评估维度到模型方法

A review of artificial intelligence-generated image quality assessment: from evaluation dimensions to model methods

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 9

参考文献 102

相关文章 3

编辑推荐

Metrics

本文评价

对比维度	基于跨模态对齐的AIGIQA方法	基于MLLM的AIGIQA方法
核心思想	以CLIP等视觉-语言模型为基础，通过学习图像与文本在共享语义空间中的对齐关系，利用特征相似度实现质量评估	以MLLM为基础，通过指令驱动与上下文建模，直接对图像质量进行语义理解与推理式评估
优点	结构简洁，计算开销较低；依赖大规模对比学习预训；易于与传统IQA框架结合	具备更强的语义理解与推理能力，可刻画复杂失真，评估结果具有更好的可解释性与灵活性
不足	主要依赖全局语义相似度，难以捕捉细粒度失真与局部伪影；缺乏显式推理能力，对复杂语义一致性问题建模不足	模型规模大，推理成本高；评估结果受prompt设计影响较大，稳定性与一致性不足

类型	方法	AGIQA-3K				AIGCIQA2023
		Quality		Alignment		Quality		Authenticity		Correspondence
		SRCC	PLCC	SRCC	PLCC	SRCC	PLCC	SRCC	PLCC	SRCC	PLCC
基于跨模态对齐的AIGIQA方法	IPCE^[85](2024)	0.884 1	0.924 6	0.769 7	0.872 5	0.864 0	0.878 8	0.809 7	0.799 8	0.797 9	0.788 7
	IP-IQA^[95](2024)	0.863 4	0.911 6	0.757 8	0.854 4	—	—	—	—	—	—
	AMFF-NET^[87](2024)	0.856 5	0.905 0	0.753 1	0.847 6	0.840 9	0.857 3	0.774 9	0.764 3	0.778 2	0.763 8
	CLIP-AGIQA^[96](2024)	0.874 7	0.919 0	—	—	0.832 4	0.860 4	—	—	—	—
	MoE-AGIQA^[89](2024)	0.875 8	0.929 4	0.823 5	0.898 5	0.872 9	0.886 0	0.793 3	0.794 1	0.764 1	0.759 0
	PromptIQA^[97](2024)	0.850 9	0.901 3	—	—	0.850 8	0.874 3	—	—	—	—
	PELA^[98](2025)	0.883 1	0.924 7	0.789 5	0.884 7	0.867 0	0.882 2	0.810 7	0.800 7	0.807 7	0.800 1
	TPMS^[86](2025)	0.890 1	0.927 0	0.773 4	0.877 3	0.862 3	0.879 8	—	—	0.786 3	0.780 6
	CIA-Net^[99](2025)	0.870 9	0.913 4	0.779 7	0.868 7	0.855 7	0.875 6	0.799 4	0.787 2	0.797 7	0.780 0
	CHPNet^[100](2026)	0.891 1	0.930 6	0.817 2	0.908 6	0.867 7	0.888 4	0.801 0	0.792 4	0.773 9	0.766 1
	DCMPLN^[101](2026)	0.887 4	0.926 2	—	—	0.850 1	0.874 7	—	—	—	—
基于MLLM的AIGIQA方法	MA-AGIQA^[91](2024)	0.870 9	0.913 6	0.772 1	0.878 5	0.865 7	0.882 1	0.813 4	0.794 9	0.750 9	0.745 5
	M3-AGIQA^[93](2025)	0.904 5	0.931 7	0.852 3	0.914 2	0.861 8	0.892 2	0.828 2	0.816 5	0.806 0	0.797 3
	AGHI-QA^[102](2025)	0.903 8	0.932 2	0.847 9	0.883 0	—	—	—	—	—	—
	SC-AGIQA^[94](2025)	0.907 0	0.936 1	—	—	0.911 3	0.915 8	—	—	—	—
	MINT-IQA^[92](2026)	0.891 9	0.932 8	0.874 9	0.928 2	0.880 1	0.887 0	0.822 9	0.812 7	0.822 6	0.805 5

[1]	包永堂, 王谟钦, 王智慧, 马光晓. 融合语义特征的全景图像质量评估[J]. 图学学报, 2026, 47(2): 332-340.
[2]	厉向东, 夏涵飞, 单逸飞, 阴凯琳, 耿卫东. 图形用户界面自动生成技术的现状与挑战[J]. 图学学报, 2024, 45(3): 409-421.
[3]	刘文军, 杨建新, 兰小平, 王波, 杨一铭, 李翠霞. 炮弹制造质量综合评估方法研究[J]. 图学学报, 2021, 42(5): 801-808.

方法	AGIQA-3K
方法	SRCC	KRCC	PLCC
CLIP^[81](2021)	0.597 2	0.459 1	0.683 9
ImageReward^[82](2023)	0.729 8	0.539 0	0.786 2
HPS^[83](2023)	0.634 9	0.458 0	0.700 0
PickScore^[33](2023)	0.697 7	0.506 9	0.763 3
StairReward^[1](2023)	0.747 2	0.555 4	0.852 9

方法	AGIQA-3K
方法	SRCC	KRCC	PLCC
CLIP^[81](2021)	0.597 2	0.459 1	0.683 9
ImageReward^[82](2023)	0.729 8	0.539 0	0.786 2
HPS^[83](2023)	0.634 9	0.458 0	0.700 0
PickScore^[33](2023)	0.697 7	0.506 9	0.763 3
StairReward^[1](2023)	0.747 2	0.555 4	0.852 9