Realistic Image Synthesis and Enhancement Using Deep Learning Architecture for Low-Resource Image Recognition Tasks
Shivam Samarendu Chaudhary
Image Synthesis, Image Enhancement, Generative Adversarial Network, YOLO, Hybrid Deep Learning, Automobile Damage Detection, Low-Resource Image Recognition, Object Detection, Data Augmentation, Deep Learning.
Image recognition systems operating in low-resource environments often face challenges arising from poor lighting, low resolution, image degradation, and limitations of image-capturing hardware. Such conditions can adversely affect the reliability of automated recognition and detection systems. In automobile inspection, these challenges become particularly important because vehicle damage such as dents, cracks, and other defects may be captured through images of varying quality and under challenging imaging conditions.
This Dissertation Phase – II presents an experimental study on a hybrid deep learning Model that integrates a Generative Adversarial Network (GAN) for realistic image synthesis and image enhancement with the You Only Look Once (YOLO) architecture for automobile damage detection. The study extends the research direction established during Dissertation Phase – I and evaluates the proposed approach using the Automobile Damage Detection dataset obtained from Roboflow Universe. The dataset comprises 6,957 annotated images belonging to fourteen automobile damage classes. The proposed Model involves image preprocessing and standardisation, GAN-based image enhancement and realistic synthetic image generation, followed by YOLO-based object detection and classification.
The developed hybrid Model was experimentally evaluated using accuracy, precision, recall, and F1-score. The results demonstrated that the hybrid YOLO–GAN model achieved an accuracy of 96%, precision of 93%, recall of 91%, and F1-score of 92%. The results demonstrate the effectiveness of integrating GAN-based image enhancement and synthesis with YOLO-based object detection for improving automobile damage recognition under challenging and low-resource imaging conditions.
The study establishes a hybrid deep learning approach for automobile damage detection and demonstrates the potential of combining generative and detection-based architectures within a unified recognition Model. The proposed approach can support applications including automobile inspection, insurance claim assessment, vehicle service inspection, and roadside damage assessment. Future work may focus on computational optimisation, lightweight architectures, and deployment in resource-constrained environments.
"Realistic Image Synthesis and Enhancement Using Deep Learning Architecture for Low-Resource Image Recognition Tasks", IJSDR - International Journal of Scientific Development and Research (www.IJSDR.org), ISSN:2455-2631, Vol.11, Issue 11, page no.b1-b39, November-2026, Available :https://ijsdr.org/papers/IJSDRTH01029.pdf
Volume 11
Issue 11,
November-2026
Pages : b1-b39
Paper Reg. ID: IJSDR_312495
Published Paper Id: IJSDRTH01029
Downloads: 00011
Research Area: Science and Technology
Country: Ahmedabad , Gujarat , India
ISSN: 2455-2631 | IMPACT FACTOR: 9.15 Calculated By Google Scholar | ESTD YEAR: 2016
An International Scholarly Open Access Journal, Peer-Reviewed, Refereed Journal Impact Factor 9.15 Calculate by Google Scholar and Semantic Scholar | AI-Powered Research Tool, Multidisciplinary, Monthly, Multilanguage Journal Indexing in All Major Database & Metadata, Citation Generator
Publisher: IJSDR(IJ Publication) Janvi Wave
