Variable-bit-rate image compression method and system, apparatus, terminal, and storage medium

Patent metadata

Field	Value
Publication number	US-12439046-B2
Application number	US-202519218104-A
Country	US
Kind code	B2
Filing date	May 23, 2025
Priority date	Dec 12, 2022
Publication date	Oct 7, 2025
Grant date	Oct 7, 2025

Abstract

The present disclosure provides a variable-bit-rate image compression method and system, an apparatus, a terminal, and a storage medium. The variable-bit-rate image compression method includes: obtaining an initial feature map from a to-be-encoded image; quantizing the initial feature map by a dead-zone quantizer; performing entropy encoding on the quantized feature map and hyper-prior information to obtain a compressed bit-stream; performing entropy decoding on the compressed bit-stream, and recovering quantized hyper-prior information and the quantized feature map; performing inverse quantization on the quantized feature map to obtain a reconstructed feature map; obtaining a reconstructed image from the reconstructed feature map; and adjusting quantization and inverse quantization parameters according to a target bit-rate or target distortion. The present disclosure provides a precise bit-rate control solution, makes the bit-rate of the compressed bit-stream better adapt to the dynamic change of a network bandwidth, and has an extremely high actual application value.

First claim

What is claimed is:

1. A variable-bit-rate image compression method, comprising:
forward mapping: performing forward mapping on a to-be-encoded image through a first decomposition transform neural network to obtain an initial feature map of the to-be-encoded image;
quantization: quantizing the initial feature map by a dead-zone quantizer to obtain a quantized feature map of the image;
entropy encoding: performing entropy encoding on the quantized feature map and quantized hyper-prior information involved in an entropy model by using the entropy model to obtain a compressed bit-stream;
entropy decoding: performing entropy decoding on the compressed bit-stream, and sequentially recovering the quantized hyper-prior information and the quantized feature map by using the entropy model;
inverse quantization: performing inverse quantization on the recovered quantized feature map to obtain a reconstructed feature map of the image;
inverse mapping: performing inverse mapping on the reconstructed feature map through a first synthesis transform neural network to obtain a reconstructed image; and
bit-rate control: adjusting quantization and inverse quantization parameters in an encoding process according to a target bit-rate or target distortion so that a bit-rate of the compressed bit-stream is close to the target bit-rate or a distortion of the reconstructed image is close to the target distortion;
the first decomposition transform neural network comprising:
a down-sampling subnetwork, wherein the down-sampling subnetwork is implemented by a convolutional neural network and is configured to transform an input image into a hidden representation; and
a reversible encoding subnetwork, wherein the reversible encoding subnetwork is configured to transform the hidden representation into the initial feature map; the reversible encoding subnetwork is obtained by hierarchical arrangement of reversible encoding units, the hierarchical arrangement comprises/layers, the i th layer (1≤i≤I) comprises 2 i−1 reversible encoding units, each of the reversible encoding units is provided with two outputs, and the two outputs of the reversible encoding unit in the i th layer are used as inputs of two reversible encoding units in the (i+1) th layer; for the unique reversible encoding unit in the first layer, an input thereof is the hidden representation; and for 2 I outputs provided in the I th layer, the initial feature map can be obtained by merging.

2. The variable-bit-rate image compression method according to claim 1 , wherein processes performed in the reversible encoding units comprise:
feature decomposition, wherein an input is divided to obtain two paths of sub-signals; and
reversible encoding, wherein for two paths of sub-signals x 1 and x 2 , reversible encoding processes thereof comprise:

𝓎
1

=

x
1

+

ℱ
⁡
(

x
2

;
θ

)

𝓎
2

=

x
2

+

𝒢
⁡
(

𝓎
1

;
μ

)

wherein F(⋅; θ) and G(⋅; μ) are both convolutional neural networks.

3. The variable-bit-rate image compression method according to claim 1 , wherein the quantization is to output the quantized feature map meeting requirements for the target bit-rate or the target distortion by adjusting a quantization step size of the dead-zone quantizer; wherein for an element y on any position in the initial feature map, a quantization output thereof is expressed as:

C
(

𝓎

;
q

,
𝓏

)

=

sign
(

𝓎

)

·

max

⁢

(

0
,

⌊

❘
"\[LeftBracketingBar]"

𝓎

❘
"\[RightBracketingBar]"

q

⁢
−
⁢

𝓏
2

+
1

⌋

)

,

wherein when y is a positive number, sign(y) is 1, when y is a negative number, sign(y) is −1, and when y is zero, sign(y) is 0; max(a, b) returns the larger one in a, b; └y┘ returns the maximum integer not greater than y; and q is the quantization step size, and z is a dead-zone rate.

4. The variable-bit-rate image compression method according to claim 1 , wherein
the entropy encoding comprises: estimating Gaussian distribution of elements on the quantized feature map by adopting the entropy model, performing arithmetic encoding on the quantized feature map, and performing arithmetic encoding on the quantized hyper-prior information by adopting structure distribution to obtain the compressed bit-stream.

5. The variable-bit-rate image compression method according to claim 1 , wherein the entropy decoding comprises: performing arithmetic decoding on the quantized hyper-prior information by adopting structure distribution, and inputting the quantized hyper-prior information to the entropy model to obtain code word distribution of the quantized feature map; and performing arithmetic decoding on the compressed bit-stream, and recovering the quantized feature map.

Assignees

• Univ Shanghai Jiaotong

Inventors

• Dai Wenrui
• Li Shaohui
• Li Chenglin
• Zou Junni
• Xiong Hongkai

Classifications

• H04N19/42

  characterised by implementation details or hardware specially adapted for video compression or decompression, e.g. dedicated software implementation (H04N19/635 takes precedence) · CPC title

• H04N19/30

  using hierarchical techniques, e.g. scalability (H04N19/63 takes precedence) · CPC title

• H04N19/184

  the unit being bits, e.g. of the compressed video stream · CPC title

• H04N19/147 Primary

  according to rate distortion criteria (rate-distortion as a criterion for motion estimation H04N19/567) · CPC title

• H04N19/13

  Adaptive entropy coding, e.g. adaptive variable length coding [AVLC] or context adaptive binary arithmetic coding [CABAC] · CPC title