Deep image compression with a unified spatial and channel context auto-regressive model

Download

10494939.pdf

Date

2022-8

Author

Uludağ, Ali Sefkan

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

125
views

100
downloads

Recently, variational auto-encoder-based compression models have gained much attention in learned image compression field due to their dimension reduction property. These models have three essential components which are encoder, decoder and entropy network. Encoder part of the VAE transforms a high dimensional space to a lower dimensional latent space, whereas the decoder serves as a reconstructing transformation. The entropy model is trained to generate the probability distributions of the latent representation to be used in arithmetic coding. End-to-end optimization is performed to minimize a rate-distortion loss composed of weighted sum of distortion loss of the decoder output and rate loss of the entropy model output. The VAE-based learned lossy image compression first introduced in Ballé (2016). Ballé (2018), improved this model by adapting a hyperprior network to increase the expression ability of spatial dependencies in the latent variable. Spatial auto-regressive and channel auto-regressive models are introduced in Minnen (2018) and Minnen (2020) respectively, to decrease the entropy of latent representation. This thesis presents a lossy image compression model which builds upon the aforementioned image compression architectures. It has been observed that spatial auto-regressive and channel auto-regressive models can be complementary and enhance the rate-distortion performance of the network. The channel auto-regressive model splits the latent representation along the channel dimension. It eliminates redundancies between channel slices by conditioning the slice being decoded to previously decoded slices. Spatial auto-regressive model captures the spatial correlations within a slice. Furthermore, it has been observed that the cumbersome spatial auto-regressive model can be parallelized by dividing the latent variables into smaller patches. Elements with the same index across patches are processed together, allowing parallel computation.

Subject Keywords

Image compression, Deep learning, Transform coding, Auto-encoder

URI

https://hdl.handle.net/11511/99476

Collections

Graduate School of Natural and Applied Sciences, Thesis

Suggestions

OpenMETU
Core

End-to-end learned image compression with conditional latent space modeling for entropy coding Yesilyurt, Aziz Berkay; Kamışlı, Fatih (2021-01-24) The use of neural networks in image compression enables transforms and probability models for entropy coding which can process images based on much more complex models than the simple Gauss-Markov models in traditional compression methods. All at the expense of higher computational complexity. In the neural-network based image compression literature, various methods to model the dependencies in the transform domain/latent space are proposed. This work uses an alternative method to exploit the dependencies o...
PROGRESSIVE COMPRESSION OF DIGITAL ELEVATION DATA USING MESHES Kose, Kivanc; Yılmaz, Erdal; ÇETİN, AHMET ENİS (2009-07-17) In this paper a new Digital Elevation Map (DEM) image compression algorithm is proposed. DEM image can be threated as a grayscale image, whose pixel values are the elevation values of the map points. The grayscale DEM image is compressed using an adaptive wavelet based image compression algorithm. The method, which is an extension of the progressive mesh compression takes advantage of the multiresolution property of the wavelets while coding the map images. This makes it possible to decode different resolut...
Deep learning-based encoder for one-bit quantization Balevi, Eren; Andrews, Jeffrey G. (2019-12-01) © 2019 IEEE.This paper proposes a deep learning-based error correction coding for AWGN channels under the constraint of one-bit quantization in receivers. An autoencoder is designed and integrated with a turbo code that acts as an implicit regularization. This implicit regularizer facilitates approaching the Shannon bound for the one-bit quantized AWGN channels even if the autoencoder is trained suboptimally, since one-bit quantization stymies ideal training. Our empirical results show that the proposed cod...
Efficient algorithms for convolutional inverse problems in multidimensional imaging Doğan, Didem; Öktem, Figen S.; Department of Electrical and Electronics Engineering (2020) Computational imaging is the process of indirectly forming images from measurements using image reconstruction algorithms that solve inverse problems. In many inverse problems in multidimensional imaging such as spectral and depth imaging, the measurements are in the form of superimposed convolutions related to the unknown image. In this thesis, we first provide a general formulation for these problems named as convolutional inverse problems, and then develop fast and efficient image reconstruction algorith...
DEEP LEARNING-BASED UNROLLED RECONSTRUCTION METHODS FOR COMPUTATIONAL IMAGING Bezek, Can Deniz; Öktem, Sevinç Figen; Department of Electrical and Electronics Engineering (2021-9-08) Computational imaging is the process of forming images from indirect measurements using computation. In this thesis, we develop deep learning-based unrolled reconstruction methods for various computational imaging modalities. Firstly, we develop two deep learning-based reconstruction methods for diffractive multi-spectral imaging. The first approach is based on plug-and-play regularization with deep denoisers whereas the second one is an end-to-end learned reconstruction based on unrolling. Secondly, we con...

Citation Formats

A. S. Uludağ, “Deep image compression with a unified spatial and channel context auto-regressive model,” M.S. - Master of Science, Middle East Technical University, 2022.