An extended context-based entropy hybrid modeling for image compression

Learned image compression with generalized octave convolution and cross-resolution parameter estimation

“…Most DNN-based image compression schemes [1,6,16,17,30,5,41,38,33] are based on the autoencoder framework to extract a low-dimensional latent representation of the input image. In [7], Ballé et al proposed a two-layer approach, which introduces the hyper layer on top of the autoencoder layer.…”

Reduced encoder time complexity using enhanced adaptive multiple transform selection in versatile video coding

“…Usage of orthogonal transforms in signal processing done in digital enhanced that led to recognizing patterns [3] and filtering using a wiener filter [4]. Noninvertible transformation enabled with orthogonal transforms that transform from pattern space to feature space with less dimensionality.…”

“…Orthogonal transforms like walsh-hadamard, discrete-fourier, slant, and Haar were considered for different applications because they have fast computing algorithms [3]- [10] for them. Karhunen-loeve transform (KLT) is optimal with respect to the following: variance distribution [3], Estimation mean-square error criterion [4], [6] and rate-distortion function [6], lacks an algorithm that enables its fast computation [3]. It is here, DCT is introduced with the fast algorithm for its computation whose performance is closer to that of KLT compared to the performance of walsh hadamard transform, discrete fourier transform, and hadamard transform.…”