Neighboring Algorithm for Visual Semantic Analysis toward GAN-Generated Pictures
Generative adversarial network (GAN)-guided visual quality evaluation means scoring GAN-propagated portraits to quantify the degree of visual distortions. In general, there are very few image- and character-evaluation algorithms generated by GAN, and the algorithm’s athletic ability is not capable....
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Wiley
2022-01-01
|
| Series: | Applied Bionics and Biomechanics |
| Online Access: | http://dx.doi.org/10.1155/2022/2188152 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1849404944079126528 |
|---|---|
| author | Lu-Ming Zhang Yichuan Sheng |
| author_facet | Lu-Ming Zhang Yichuan Sheng |
| author_sort | Lu-Ming Zhang |
| collection | DOAJ |
| description | Generative adversarial network (GAN)-guided visual quality evaluation means scoring GAN-propagated portraits to quantify the degree of visual distortions. In general, there are very few image- and character-evaluation algorithms generated by GAN, and the algorithm’s athletic ability is not capable. In this article, we proposed a novel image ranking algorithm based on the nearest neighbor algorithm. It can obtain automatic and extrinsic evaluation of GAN procreate images using an efficient evaluation technique. First, with the support of the artificial neural network, the boundaries of the variety images are extracted to form a homogeneous portrait candidate pool, based on which the comparison of product copies is restricted. Subsequently, with the support of the K-nearest neighbors algorithm, from the unified similarity candidate pool, we extract the most similar concept of K-Emperor to the generated portrait and calculate the portrait quality score accordingly. Finally, the property of generative similarity that produced by the GAN models are trained on a variety of classical datasets. Comprehensive experimental results have shown that our algorithm substantially improves the efficiency and accuracy of the natural evaluation of pictures generated by GAN. The calculated metric is only 1/9–1/28 compared to the other methods. Meanwhile, the objective evaluation of the GAN and human consistency has increased by more than 80% in line with human visual perception. |
| format | Article |
| id | doaj-art-99b4692b5d6849fca91aa151bac34822 |
| institution | Kabale University |
| issn | 1754-2103 |
| language | English |
| publishDate | 2022-01-01 |
| publisher | Wiley |
| record_format | Article |
| series | Applied Bionics and Biomechanics |
| spelling | doaj-art-99b4692b5d6849fca91aa151bac348222025-08-20T03:36:48ZengWileyApplied Bionics and Biomechanics1754-21032022-01-01202210.1155/2022/2188152Neighboring Algorithm for Visual Semantic Analysis toward GAN-Generated PicturesLu-Ming Zhang0Yichuan Sheng1Key Laboratory of Crop Harvesting Equipment Technology of Zhejiang ProvinceKey Laboratory of Crop Harvesting Equipment Technology of Zhejiang ProvinceGenerative adversarial network (GAN)-guided visual quality evaluation means scoring GAN-propagated portraits to quantify the degree of visual distortions. In general, there are very few image- and character-evaluation algorithms generated by GAN, and the algorithm’s athletic ability is not capable. In this article, we proposed a novel image ranking algorithm based on the nearest neighbor algorithm. It can obtain automatic and extrinsic evaluation of GAN procreate images using an efficient evaluation technique. First, with the support of the artificial neural network, the boundaries of the variety images are extracted to form a homogeneous portrait candidate pool, based on which the comparison of product copies is restricted. Subsequently, with the support of the K-nearest neighbors algorithm, from the unified similarity candidate pool, we extract the most similar concept of K-Emperor to the generated portrait and calculate the portrait quality score accordingly. Finally, the property of generative similarity that produced by the GAN models are trained on a variety of classical datasets. Comprehensive experimental results have shown that our algorithm substantially improves the efficiency and accuracy of the natural evaluation of pictures generated by GAN. The calculated metric is only 1/9–1/28 compared to the other methods. Meanwhile, the objective evaluation of the GAN and human consistency has increased by more than 80% in line with human visual perception.http://dx.doi.org/10.1155/2022/2188152 |
| spellingShingle | Lu-Ming Zhang Yichuan Sheng Neighboring Algorithm for Visual Semantic Analysis toward GAN-Generated Pictures Applied Bionics and Biomechanics |
| title | Neighboring Algorithm for Visual Semantic Analysis toward GAN-Generated Pictures |
| title_full | Neighboring Algorithm for Visual Semantic Analysis toward GAN-Generated Pictures |
| title_fullStr | Neighboring Algorithm for Visual Semantic Analysis toward GAN-Generated Pictures |
| title_full_unstemmed | Neighboring Algorithm for Visual Semantic Analysis toward GAN-Generated Pictures |
| title_short | Neighboring Algorithm for Visual Semantic Analysis toward GAN-Generated Pictures |
| title_sort | neighboring algorithm for visual semantic analysis toward gan generated pictures |
| url | http://dx.doi.org/10.1155/2022/2188152 |
| work_keys_str_mv | AT lumingzhang neighboringalgorithmforvisualsemanticanalysistowardgangeneratedpictures AT yichuansheng neighboringalgorithmforvisualsemanticanalysistowardgangeneratedpictures |