Convolutional Deep Colorization for Image Compression: A Color Grid Based Approach

TL;DR

The paper addresses image compression by retaining a grid of color pixels and using a convolutional recolorization network (Crayon) derived from XiaoNet. It processes images in LAB color space, training with mean squared error loss and ADAM optimization to recolor AB channels from the color grid. It shows a trade-off where the compression upper bound is $relativeSize = \frac{1}{3} + \frac{1}{n^2}$ and the best balance occurs near $n=20$, yielding competitive CSIM and PSNR with meaningful storage savings. The work demonstrates promise for practical compression with limited artifacts and suggests directions for adaptive color-pixel selection and architecture generalization.

Abstract

The search for image compression optimization techniques is a topic of constant interest both in and out of academic circles. One method that shows promise toward future improvements in this field is image colorization since image colorization algorithms can reduce the amount of color data that needs to be stored for an image. Our work focuses on optimizing a color grid based approach to fully-automated image color information retention with regard to convolutional colorization network architecture for the purposes of image compression. More generally, using a convolutional neural network for image re-colorization, we want to minimize the amount of color information that is stored while still being able to faithfully re-color images. Our results yielded a promising image compression ratio, while still allowing for successful image recolorization reaching high CSIM values.

Figures (3)

• Figure 1: Example decolorization process on low resolution image with n=3.
• Figure 2: Image quality and relative compression size of images for varied color grid spacing values.
• Figure 3: Image recolorization with varying levels of color retention. As n increases, images look less like the original image.