Image Processing Model Selection for Mixed Pixel Value Ranges
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing methods using machine learning models struggle with precision when handling images in different formats such as JPEG and HEIF, due to differences in representable value ranges and image capturing modes, leading to suboptimal results when applying models trained on one format to images in another.
Innovation Solution
An image processing method that selects a machine learning model based on range information, including image format and capturing mode, and performs normalization and denormalization using specific constants to ensure accurate processing, regardless of the image format.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single machine learning model is used for all image formats, then device complexity is reduced, but manufacturing precision deteriorates due to inability to handle different value ranges
Solution Approach 1:
The patent changes the parameter of pixel value range to match the input image format by selecting different normalization constants. When the input image is in JPEG format (0-255 range), the model uses normalization constant 255; when in HEIF format (0-1023 range), it uses normalization constant 1023. This parameter adaptation allows a single model to process multiple formats with high precision without increasing structural complexity.
2Manufacturing precision
If multiple machine learning models are maintained for different formats, then manufacturing precision is improved, but device complexity increases
Solution Approach 1:
The patent makes a single machine learning model universal by enabling it to handle multiple image formats through dynamic parameter adjustment. The model structure remains unchanged, but it adapts to different formats (JPEG, HEIF, and others) by selecting appropriate normalization constants based on the input image's pixel value range. This multi-functionality approach achieves format-agnostic processing without maintaining separate models.
3Manufacturing precision
If normalization constants are adjusted based on image format, then manufacturing precision is improved, but ease of operation deteriorates due to additional selection steps
Solution Approach 1:
The system performs self-service by automatically detecting the input image format and selecting the appropriate normalization constant without requiring manual intervention. The image processing apparatus autonomously determines whether the input image uses JPEG (0-255), HEIF (0-1023), or other formats and adjusts the normalization parameter accordingly, making the precision improvement transparent to the user.
4Ease of operation
If a machine learning model trained on JPEG images is applied to HEIF images, then ease of operation is improved, but manufacturing precision deteriorates due to range mismatch
Solution Approach 1:
The patent resolves the range mismatch problem by dynamically changing the normalization parameter based on the input image format. When processing HEIF images (0-1023 range) with a model trained on JPEG images, the system adjusts the normalization constant to 1023, ensuring the input data distribution matches the training conditions. This parameter adaptation maintains processing precision across different formats while preserving model applicability.
Data Source
AI summary
A method and the like for performing image processing with higher precision on various image data using a machine learning model are provided. The method includes obtaining an input image and range information about pixel values of the input image, selecting at least one machine learning model from among a plurality of machine learning models based on the range information, and generating an estimated image by inputting the input image to the selected machine learning model. Alternatively, the method includes obtaining an input image and range information about the input image, and generating an estimated image by inputting the input image and the range information to a machine learning model.