Image Compression: Sparse Coding vs. Bottleneck Autoencoders
share
Bottleneck autoencoders have been actively researched as a solution to image compression tasks. In this work, we explore the ability of sparse coding to improve reconstructed image quality for the same degree of compression. We observe that sparse image compression provides qualitatively superior visual quality of reconstructed images but has lower values of PSNR and SSIM compared to bottleneck autoencoders. We hypothesized that there should be another evaluational criterion to support our subjective observations. To test this hypothesis, we fed reconstructed images from both the bottleneck autoencoder and sparse coding into a DCNN classifier and discovered that the images reconstructed from the sparse coding compression obtained on average 1.5% higher classification accuracy compared to bottleneck autoencoders, implying that sparse coding preserves more content-relevant information.
READ FULL TEXT