Convolutional Neural Network Image Compression System
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The manual setting of filtering parameters in image compression and decompression systems using wavelet transforms is complex and error-prone, making it difficult to achieve optimal compression rates.
Innovation Solution
An image compression system utilizing convolutional neural networks to predict and update images, determining discrepant features, and superposing them to produce a compressed image, with a corresponding decompression system to reverse the process, thereby optimizing filtering parameters automatically.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If manual setting of filtering parameters is used in wavelet transform-based image compression, then the system structure remains simple, but the compression rate cannot be optimized and the parameter setting becomes complex and error-prone
Solution Approach 1:
The system employs a training device that automatically trains the filtering units to determine optimal filtering parameters through self-learning from training images. The training device calculates loss values based on compression quality and iteratively adjusts parameters without manual intervention, enabling the system to serve itself in optimizing compression performance.
Solution Approach 2:
The invention transforms the static manual parameter setting into dynamic parameter optimization by continuously adjusting filtering parameters during the training process. The training device modifies parameters based on loss function feedback, achieving optimal compression rates adaptively rather than relying on fixed manually-set values.
2Productivity
If convolutional neural networks are introduced for automatic parameter optimization, then the compression rate is improved, but the device complexity increases
Solution Approach 1:
The training device merges the compression engine and training functions into a unified system. The same convolutional neural network modules used for compression are also utilized for training, with the training device orchestrating both processes. This integration reduces overall system complexity compared to having separate independent systems.
Solution Approach 2:
The filtering units serve dual purposes: they perform image compression during normal operation and are simultaneously trained to optimize parameters through the training device. The convolutional neural network modules function both as compression operators and as learnable parameters, eliminating the need for separate parameter optimization hardware.
3Manufacturing precision
If multiple convolutional neural network modules are used for prediction and update, then the compression quality is improved, but the computational complexity increases
Solution Approach 1:
The training device performs preliminary training of the convolutional neural network modules before actual compression operations. By pre-optimizing the filtering parameters through training on training images, the system reduces the computational burden during runtime compression, as the modules are already tuned for optimal performance rather than requiring extensive real-time computation.
Data Source
AI summary
The disclosure relates to an image compression system, an image decompression system, a training method and device, as well as a display device. In the image compression system, convolutional neural network modules are used to complete the update and prediction processes. As such, the weight of each filtering unit in the convolutional neural network module can be trained in order to provide the corresponding image compression system with a better compression rate, thereby reducing the difficulty in setting the filtering parameters of the image compression unit and the image decompression unit.