Single Neural Network for Mobile Image Parameter Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image enhancement technologies on mobile devices are limited by processing power and memory, requiring multiple neural networks for each parameter and resulting in inefficient processing and high computational costs, making it impractical for real-time aesthetic enhancements.
Innovation Solution
A single neural network is trained to evaluate multiple control parameters, using convolutional layers and parameter heads to calculate offset values for each parameter, allowing for efficient image enhancement with reduced processing requirements and faster processing times.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple neural networks are used to evaluate each control parameter separately, then measurement precision of parameter evaluation is improved, but device complexity and processing power requirements increase significantly
Solution Approach 1:
The patent combines multiple separate neural networks into a single unified neural network that evaluates multiple control parameters simultaneously. The network architecture includes a shared feature extraction backbone followed by separate parameter-specific evaluation heads, allowing one network to perform the work of multiple networks while maintaining parameter-specific evaluation accuracy
Solution Approach 2:
The unified neural network is designed with multi-functionality to handle multiple control parameters (e.g., brightness, contrast, saturation, sharpness) within a single system. The network accepts image data and outputs evaluations for multiple parameters through shared layers and specialized heads, making the system universal across different parameter types
2Manufacturing precision
If multiple image variations are generated to find optimal aesthetic enhancements, then image enhancement quality is improved, but processing time and computational power increase
Solution Approach 1:
The system performs preliminary evaluation by using the unified neural network to assess multiple control parameters simultaneously on the original image, identifying optimal parameter adjustments before generating any image variations. This preliminary parameter optimization guides subsequent image generation, reducing the need for extensive trial-and-error with multiple variations
Solution Approach 2:
The patent replaces the mechanical approach of generating multiple image variations and manually or algorithmically selecting the best one with an intelligent system. The unified neural network directly predicts optimal parameter values based on image content and aesthetic principles, substituting brute-force variation generation with intelligent parameter estimation
3Measurement precision
If high resolution images are used for parameter optimization, then measurement precision of aesthetic evaluation is improved, but processing power and memory requirements increase
Solution Approach 1:
The neural network architecture is segmented into a shared feature extraction backbone that processes image data at optimized resolutions, and separate parameter-specific evaluation heads that operate on the extracted features. This segmentation allows the computationally intensive feature extraction to be performed once on potentially downscaled images, while parameter evaluation operates on the compact feature representations, reducing overall processing power requirements while maintaining accuracy
Data Source
AI summary
A method and system of generating an adjustment parameter value for a control parameter to enhance a new image, which includes configuring a neural network, trained to restore image quality for a derivative image, to that of an earlier version of the derivative image, to generate as an output the adjustment parameter value, for the control parameter in response to input of data derived from the new image, and changing a control parameter of the new image, by generating the adjustment parameter value by calculating an inverse of the output value, and applying the adjustment parameter value to the control parameter of the new image so as to generate an enhanced image.