Optimization of Fourier Space Filter for Chest X-ray Image Compression in Deep Learning Training

Xinyi Li      Yuhao Yan     

xl233@duke.edu     yy263@duke.edu    

This project proposed a Fourier-transform-based (FT-based) image compression filter for Chest X-ray images in the Deep Learning training to distinguish normal and pneumonia images. The FT filter was initialized with a Butterworth low-pass filter, and was optimized during network training of a reduced ResNet. The resulting FT compression layer could compress images to 1/4 of their original sizes and slightly improve prediction AUC. Please refer to the following links for more information about this project.

Paper: Paper PDF

Code and Data: The link to code (right click to download): jupyter notebook The link to data: Chest X-ray Pneumonia