Subject

Digital imaging has led to an exponential growth in visual data, necessitating efficient image compression techniques to manage storage and bandwidth requirements. Traditional image compression methods focus on reducing redundancy in pixel data without considering the semantic content of the images. This thesis explores advanced semantic image compression techniques that leverage deep learning models to achieve higher compression rates while preserving perceptually important features. By incorporating semantic understanding into the compression process, this research seeks to significantly improve the efficiency and effectiveness of image compression, with potential applications in various fields such as multimedia storage, transmission, and medical imaging.

Kind of work

The objectives of the thesis are:
1. Literature Review: Conduct a comprehensive review of existing image compression techniques, including traditional methods (JPEG, PNG) and modern approaches (deep learning-based compression).
2. Deep Learning Models: Investigate state-of-the-art deep learning models for image analysis and compression, such as convolutional neural networks (CNNs) and generative adversarial networks (GANs).
3. Semantic Segmentation: Implement semantic segmentation to identify and prioritize important regions within images for higher quality preservation during compression.
4. Model Development: Develop and train a novel deep learning model that integrates semantic segmentation with image compression techniques.
5. Performance Evaluation: Evaluate the performance of the proposed method in terms of compression ratio, image quality, and computational efficiency. Compare with existing compression standards.
6. Applications: Explore potential applications of the developed compression technique in areas like multimedia, and healthcare

Framework of the Thesis

Expected Outcomes:
- A novel deep learning-based semantic image compression model that outperforms traditional methods in preserving image quality while achieving higher compression ratios.
- Comprehensive evaluation metrics demonstrating the efficacy of the proposed method.
- Potential use cases and applications demonstrating the practical benefits of semantic image compression.

References:
A list of recent papers on image compression, semantic segmentation, and deep learning techniques is given below:
- Zhang, Pingping, Shiqi Wang, Meng Wang, Jiguo Li, Xu Wang, and Sam Kwong. "Rethinking semantic image compression: Scalable representation with cross-modality transfer." IEEE Transactions on Circuits and Systems for Video Technology (2023).
- Chang, J., Zhang, J., Li, J., Wang, S., Mao, Q., Jia, C., ... & Gao, W. (2023). Semantic-aware visual decomposition for image coding. International Journal of Computer Vision, 131(9), 2333-2355.
- Feng, R., Gao, Y., Jin, X., Feng, R., & Chen, Z. (2023). Semantically structured image compression via irregular group-based decoupling. In Proceedings of the IEEE/CVF International Conference on Computer Vision (pp. 17237-17247).

This thesis will contribute to the advancement of image compression technologies by integrating semantic analysis, offering a pathway to more efficient and intelligent handling of visual data.

Number of Students

1 or 2

Expected Student Profile

Good knowledge of machine learning, deep learning and image processing. Good knowledge of Python (including PyTorch)