Learning to Inpaint for Image Compression
This work addresses efficient image compression for applications requiring high-quality storage, presenting incremental improvements over existing methods.
The paper tackles image compression by introducing a multi-stage progressive encoder that predicts original data from residuals and uses inpainting to reduce stored information, achieving over 60% reduction in file size with similar quality compared to a baseline.
We study the design of deep architectures for lossy image compression. We present two architectural recipes in the context of multi-stage progressive encoders and empirically demonstrate their importance on compression performance. Specifically, we show that: (a) predicting the original image data from residuals in a multi-stage progressive architecture facilitates learning and leads to improved performance at approximating the original content and (b) learning to inpaint (from neighboring image pixels) before performing compression reduces the amount of information that must be stored to achieve a high-quality approximation. Incorporating these design choices in a baseline progressive encoder yields an average reduction of over $60\%$ in file size with similar quality compared to the original residual encoder.