Object-Type AI Model Selection for Adaptive Image Upscaling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing neural network models applied universally to all inputs in electronic apparatuses result in suboptimal performance due to varying input characteristics, leading to side effects like jagging or blurring, and methods to address these issues are inefficient in resource-constrained environments.
Innovation Solution
An electronic apparatus equipped with a memory storing multiple AI models and a processor that identifies the type of an input image, down-scales it if necessary, and selects the appropriate AI model for processing based on the identified type, enhancing processing performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single neural network model is applied to all input images, then device complexity is reduced, but manufacturing precision deteriorates due to side effects like jagging or blurring in specific image types
Solution Approach 1:
The patent divides the single neural network model into multiple specialized models (first neural network model for general images, second neural network model for text-containing images, third neural network model for object-containing images). Each model is trained specifically for its designated image type to eliminate side effects like jagging or blurring that occur when a general model processes specialized content.
Solution Approach 2:
The patent applies different processing qualities to different regions/types of images. Text-containing images receive processing from the second neural network model optimized for text preservation, object-containing images receive processing from the third neural network model optimized for object recognition, while general images use the first model. This ensures each image type receives the appropriate level and type of processing quality.
2Manufacturing precision
If multiple neural network models are stored for different image types, then manufacturing precision is improved, but device complexity increases
Solution Approach 1:
The patent introduces a determination module as an intermediary that automatically identifies the type of input image and selects the appropriate neural network model. This mediator handles the complexity of managing multiple models by providing a simple interface: the determination module analyzes the input image characteristics and routes it to the correct specialized model, shielding the user from model management complexity.
Solution Approach 2:
The patent changes the parameter of model selection based on image type characteristics. The determination module detects parameters such as the presence of text or objects in the input image and dynamically selects which neural network model to apply. This parameter-based selection mechanism manages the complexity of having multiple models by making selection automatic and condition-based rather than manual.
3Measurement precision
If image processing is performed without down-scaling, then measurement precision is maintained, but loss of time increases due to longer processing duration
Solution Approach 1:
The patent performs down-scaling as a preliminary action before feeding the image to the neural network model for object identification. By reducing the image resolution beforehand, the overall processing time is reduced while the neural network model maintains sufficient accuracy for identifying object types. This preliminary preprocessing step prepares the data in an optimal format for subsequent processing.
Data Source
AI summary
An electronic apparatus is provided. The electronic apparatus includes a memory storing information on a first artificial intelligence model for identifying a type of an object included in an image and information on a plurality of second artificial intelligence models for upscaling the image and a processor connected to the memory and configured to control the electronic apparatus, and the processor is configured to input an input image to the first artificial intelligence model and identify a type of an object included in the input image, and upscale the input image by inputting the input image to one of the plurality of second artificial intelligence models based on the identified type of the object.