Image Restoration Model Using Text Probability Maps for Character Clarity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing technologies struggle to effectively restore low-resolution images with reduced semantic bias, particularly in recognizing and enhancing characters, leading to inaccuracies and reduced clarity in distorted images.
Innovation Solution
An image restoration model is employed that includes a sub-model trained with masked attention scores, an encoder for feature extraction, a fusion layer for combining text probability maps and feature information, and a decoder to generate high-resolution images, utilizing a specified masking ratio to reduce semantic bias.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing image processing technologies are used to restore low-resolution images, then the restoration process can be performed, but the character recognition accuracy and image clarity are insufficient due to semantic bias
Solution Approach 1:
The image restoration model is segmented into distinct functional modules: encoder for feature extraction, sub-model for text probability map generation, fusion layer for combining features, and decoder for high-resolution reconstruction. This segmentation allows each module to specialize in specific tasks, improving overall character recognition accuracy while reducing semantic bias in the restoration process
Solution Approach 2:
A text probability map is introduced as an intermediary element between the encoder and decoder. This intermediate representation captures character-specific information and guides the restoration process, enabling more accurate character reconstruction while reducing semantic bias that would otherwise be present in direct restoration approaches
2Manufacturing precision
If conventional image restoration methods are applied, then the processing can be completed, but the clarity and visibility of distorted characters remain insufficient
Solution Approach 1:
The model applies local quality enhancement by generating a text probability map that provides character-specific guidance during restoration. This allows different regions of the image to be restored with appropriate detail levels, particularly enhancing character regions while maintaining overall image quality, thereby improving both restoration quality and character visibility
Solution Approach 2:
The restoration process is extended into an additional dimensional space by incorporating text probability maps as a separate channel of information. This multi-dimensional approach allows the model to simultaneously process spatial image data and character-specific probability data, improving both restoration quality and character detectability
3Measurement precision
If simple image restoration models are used, then the computational complexity is low, but the ability to reduce semantic bias and improve character recognition is limited
Solution Approach 1:
The image restoration model is designed with multi-functionality, where the encoder-submodel-fusion-decoder architecture serves multiple purposes: feature extraction, text probability generation, feature fusion, and high-resolution reconstruction. This universal design improves character recognition accuracy without requiring entirely separate systems, balancing functionality with manageable complexity
Data Source
AI summary
According to an embodiment, an electronic device obtains an input image of a first resolution that includes one or more characters. The electronic device, using the input image, performs training of an image restoration model including a sub model trained to output a text probability map representing the one or more characters associated with the input image, an encoder configured to extract feature information from the input image, a fusion layer configured to combine the text probability map and the feature information, and a decoder connected to the fusion layer and for generating an output image with a second resolution higher than the first resolution. The sub model is trained through one or more masked attention scores obtained by applying a specified masking ratio for a different single character selected among the one or more characters.