Guided-colorization-based color image coding

Abstract

This paper proposes guided-colorization-based image coding that utilizes a guide image for encoding chroma components. Up to now, several colorization-based image coding techniques have been proposed. In those techniques, the luma component is encoded by a standard encoder, while the two chroma components by colorization. Namely, the encoder selects representative chroma samples and the decoder recovers the entire chroma components from the representative chroma samples and the luma samples by solving optimization. The colorization relies on the assumption that chroma samples in a local region have a similar distribution that the corresponding luminance samples have. However, since the assumption often fails for many color images, it cannot recover the chroma components accurately. In this work, we utilize guide chroma components of a similar color image for recovering the chroma components. Such image can be found from, for example, users' photo storages. For more efficient guided colorization, we introduce global kernels which enable accurate chroma recovery from fewer representative chroma samples required in conventional methods. We perform an evaluation in lossy color image coding and show the effectiveness of the proposed method.