Image Compression Feature Gains for Single-Model Bit Rate Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image encoding methods based on deep convolutional networks are limited in that a single trained model can only output one encoding result for a specific type of input image, making it difficult to achieve the desired compression bit rate.
Innovation Solution
The proposed image processing method involves obtaining a target compression bit rate and processing corresponding feature values based on target gain values to achieve compression bit rate control within a single compression model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a single trained model is used for image encoding, then the model structure is simple and easy to deploy, but the compression bit rate cannot be controlled according to actual requirements
Solution Approach 1:
The patent introduces dynamic gain values that can be adjusted according to different compression bit rate requirements. The gain values are applied to feature maps in the encoding process, allowing the same model structure to adapt to different compression ratios by changing the gain parameters rather than changing the model itself.
Solution Approach 2:
The patent changes the parameter space by introducing gain values as controllable parameters. Instead of training multiple models for different bit rates, the system uses a single model with adjustable gain parameters that modify the feature map processing, enabling flexible bit rate control through parameter adjustment.
2Manufacturing precision
If different models are trained for different compression bit rates, then the compression bit rate control is precise, but the device complexity and number of models increase
Solution Approach 1:
The patent creates a universal model that can handle multiple compression bit rates through the introduction of gain values. The same encoding model serves multiple functions by adjusting the gain parameters, eliminating the need for separate models for different bit rates while maintaining precise control.
Solution Approach 2:
The system achieves precise bit rate control by changing parameters (gain values) rather than changing the model structure. The gain values modulate the feature map processing to achieve different compression levels, providing precise control with a single model.
3Productivity
If conventional encoding methods are used, then the encoding process is fast and simple, but the encoding effect for different compression requirements is insufficient
Solution Approach 1:
The patent maintains fast encoding by using a single model with dynamic gain adjustment. The gain values are applied during the forward pass of the encoding process, adding minimal computational overhead while enabling adaptation to different compression requirements and improving encoding effectiveness.
Data Source
AI summary
An image processing method and apparatus are provided. The method includes: obtaining an image; performing feature extraction on the image to obtain at least one first feature map, wherein the at least one first feature map includes N first feature values, and N is a positive integer; obtaining a target compression bit rate which corresponds to M target gain values, each target gain value corresponds to one first feature value, and M is a positive integer less than or equal to N; respectively processing corresponding first feature values based on the M target gain values to obtain M second feature values; and performing quantization and entropy encoding on at least one processed first feature map to obtain encoded data, wherein the at least one processed first feature map includes the M second feature values. According to the application, compression bit rate control is implemented in a same compression model.