EGIC: Enhanced Low-Bit-Rate Generative Image Compression Guided by Semantic Segmentation

We introduce EGIC, a novel generative image compression method that allows traversing the distortion-perception curve efficiently from a single model. Specifically, we propose an implicitly encoded variant of image interpolation that predicts the residual between a MSE-optimized and GAN-optimized decoder output. On the receiver side, the user can then control the impact of the residual on the GAN-based reconstruction. Together with improved GAN-based building blocks, EGIC outperforms a wide-variety of perception-oriented and distortion-oriented baselines, including HiFiC, MRIC and DIRAC, while performing almost on par with VTM-20.0 on the distortion end. EGIC is simple to implement, very lightweight (e.g. 0.18x model parameters compared to HiFiC) and provides excellent interpolation characteristics, which makes it a promising candidate for practical applications targeting the low bit range.

Differential Invariants under Gamma Correction

This paper presents invariants under gamma correction and similarity transformations. The invariants are local features based on differentials which are implemented using derivatives of the Gaussian. The use of the proposed invariant representation is shown to yield improved correlation results in a template matching scenario.

A Differential Invariant for Zooming

This paper presents an invariant under scaling and linear brightness change. The invariant is based on differentials and therefore is a local feature. Rotationally invariant 2-d differential Gaussian operators up to third order are proposed for the implementation of the invariant. The performance is analyzed by simulating a camera zoom-out.

A Linear Shift Invariant Multiscale Transform

This paper presents a multiscale decomposition algorithm. Unlike standard wavelet transforms, the proposed operator is both linear and shift invariant. The central idea is to obtain shift invariance by averaging the aligned wavelet transform projections over all circular shifts of the signal. It is shown how the same transform can be obtained by a linear filter bank.