Advanced Image Enhancement and Data Recovery: Superresolution Techniques and Missing Data Handling
This project focuses on enhancing image resolution and reconstructing images with missing data using advanced machine learning techniques. Specifically, it applies superresolution to enhance image quality and employs Random Fourier Features (RFF) combined with linear regression for image reconstruction. The project involves both qualitative and quantitative analyses to evaluate the performance of these methods. Key tasks include superresolution enhancement, measuring reconstruction accuracy with various metrics, and handling images with different levels of missing data
-
Superresolution Enhancement : Performing Superresolution on the image to enhance its resolution by factor
2. Displaying a qualitative comparison of original and reconstructed image. -
Quantitative Comparison of Superresolution : The above only helps us with a qualitative comparison. Let us now do a quantitative comparison. First read this article: https://github.com/sgrvinod/a-PyTorch-Tutorial-to-Super-Resolution
- Start with a 400x400 image (ground truth high resolution).
- Resize it to a 200x200 image (input image)
- Use
RFF+Linear regressionto increase the resolution to 400x400 (predicted high resolution image) - Compute the following metrics:
RMSEon predicted v/s ground truth high resolution image- `Peak SNR``
-
Completing Image with Random Missing Data:
- Applying RFF to complete the image with 10%, 20%, and so on up to 90% of its data missing randomly.
- Randomly remove portions of the data, train the model on the remaining data, and predict on the entire image.
- Displaying the reconstructed images for each missing data percentage and providing the metrics calculated above.