AI Learning Model for Image Contrast Ratio Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for training learning models to enhance image contrast ratios are inconvenient due to the need for manual generation of high contrast ratio images and increased complexity and memory usage with large data sizes.
Innovation Solution
An electronic device with processors and memory that predicts an inverse function of a monotonically increasing function to decrease image contrast ratio, using a learning model trained with a target image having a high contrast ratio, and generates training images by applying the target image to multiple monotonically increasing functions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If manual generation of high contrast ratio images is used for training, then training data quality is improved, but operation convenience deteriorates
Solution Approach 1:
The patent applies preliminary action by pre-generating multiple monotonically increasing functions with different parameters before the actual training process. These functions are prepared in advance to automatically generate training data pairs (low contrast ratio images and their corresponding high contrast ratio images) without requiring manual intervention during training, thus resolving the contradiction between training data quality and operation convenience
Solution Approach 2:
The system implements self-service by using the learned monotonically increasing function to automatically generate training data pairs. The learning model trains itself by generating its own training data through applying the function and its inverse function to input images, eliminating the need for manual high contrast ratio image generation and improving operational convenience while maintaining training data quality
2Measurement precision
If large data size is used for training, then model accuracy is improved, but memory usage increases
Solution Approach 1:
The patent applies dynamics by making the training data generation process dynamic rather than static. Instead of using fixed large datasets that consume significant memory, the system dynamically generates training data pairs on-demand by applying monotonically increasing functions with varying parameters to input images. This allows the model to achieve good accuracy with smaller memory footprint by generating diverse training examples through parameter variation
Solution Approach 2:
The system uses parameter changes by varying the parameters of monotonically increasing functions to generate diverse training data pairs. By changing function parameters (such as contrast enhancement strength, brightness adjustment levels), the system creates multiple variations of training images without needing to store large amounts of pre-collected data, thus improving model accuracy while reducing memory usage
Data Source
Figure 1~3
Figure 4~5b
Figure 6~7b
AI summary
An electronic device is provided. The electronic device may include at least one processor, and at least one memory. The at least one processor may be configured to execute the instructions stored in the at least one memory, to: acquire an input image, predict an inverse function of a monotonically increasing function for decreasing an image contrast ratio by applying the input image to a learning model trained by using an artificial intelligence algorithm, and acquire an output image based on the input image and the predicted inverse function of the monotonically increasing function. The learning model may be trained to predict the inverse function of the monotonically increasing function for decreasing the an image contrast ratio by using a training image generated by applying a target image having a high contrast ratio to the monotonically increasing function and the inverse function of the monotonically increasing function as training data.