CNN-Guided Image Encoding for Single-Pass Quality and Compression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image encoding methods face challenges in achieving a balance between visual quality and compression efficiency, often requiring iterative processes that are resource-intensive and time-consuming, and result in sub-optimal outcomes due to inconsistent application of compression settings across different images and codecs.
Innovation Solution
A non-iterative image encoding process using a pre-trained Convolutional Neural Network (CNN) to predict optimal encoder settings for a single encoding iteration, ensuring the generated image meets target visual-quality and/or byte-size thresholds by analyzing the input image and selecting the best encoding parameters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If iterative encoding processes are used to achieve target visual quality and compression ratio, then encoding accuracy is improved, but processing time and computational resources increase significantly
Solution Approach 1:
The patent applies preliminary action by using a pre-trained CNN model to predict optimal encoding parameters before actual encoding occurs. The system analyzes the input image, predicts the best encoder and parameters that will achieve target quality and compression, then directly applies those parameters in a single encoding pass without iteration.
Solution Approach 2:
The patent replaces the mechanical iterative encoding system with an AI-based prediction system. Instead of mechanically trying multiple encoders and parameters iteratively, the system uses a trained neural network to predict the optimal configuration, substituting computational iteration with intelligent prediction.
2Manufacturing precision
If multiple interim compressed versions are generated and evaluated, then optimal compression settings are achieved, but computational complexity and resource consumption increase
Solution Approach 1:
The patent extracts the essential information needed for optimization by using the CNN model to predict outcomes directly from the input image, without generating multiple interim compressed versions. Only the optimal single version is produced, extracting the best result without the complexity of evaluating multiple candidates.
Solution Approach 2:
The patent uses the pre-trained CNN model as a virtual copy that has learned from multiple training examples. This model copy can predict the optimal encoding parameters without actually creating multiple physical compressed versions, replacing the need for multiple real encodings with a single prediction-based approach.
3Stability of the object's composition
If different encoding schemes are applied to ensure consistent quality across images, then visual quality consistency is improved, but encoding speed decreases due to multiple trials
Solution Approach 1:
The patent changes the approach from adjusting encoding parameters iteratively to predicting optimal parameters directly. The CNN model has learned the relationship between input images and optimal encoding parameters, allowing it to directly output the correct settings without iterative adjustment, thus maintaining consistency while improving speed.
Data Source
AI summary
System, device, and method for improved image encoding that non-iteratively targets and achieves a visual-quality threshold and a compression efficiency threshold. A system receives an input image intended for compression; and applies a pre-trained Convolutional Neural Network (CNN) Engine to predict what would be (i) compression ratio values and (ii) visual-quality score values, for a resulting compressed image that would be encoded or transcoded from said input image, if said input image would be compressed via a particular Image Encoder from a pool of available Image Encoders by using a particular Encoder Settings from a pool of Image Encoder Settings. Based on CNN-based predictions generated by the pre-trained CNN Engine, the system selects a single combination of Image Encoder with Encoder Setting, and performs a single compression of said input image using said single combination of Image Encoder with Encoder Setting to generate an optimally compressed image.