Image Style Transfer With Quality-Aware Preprocessing and Color Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image style transfer methods struggle with maintaining image quality, such as object line representation, color uniformity, and sharpness, especially in low-contrast or blurry images, requiring significant time and empirical improvements.
Innovation Solution
An electronic apparatus equipped with neural network models for pre-processing and post-processing to enhance sharpness, contrast, and brightness, and a control method to manage color differences, ensuring optimal image quality during style transfer.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If style transfer is performed on low-quality images (blurry, low contrast, small scale), then the style transfer effect can be applied, but the line representation disappears or becomes distorted and image quality degrades
Solution Approach 1:
The patent applies image preprocessing operations (deblurring, contrast enhancement, upscaling) before style transfer to improve input image quality. This preliminary action ensures that line representations are preserved and not distorted during the subsequent style transfer process, resolving the contradiction between style transfer applicability and line representation quality.
Solution Approach 2:
The patent introduces an intermediary image processing stage between the original image and the style transfer process. This intermediary stage includes quality assessment and preprocessing operations that prepare the image for style transfer, ensuring both style transfer applicability and maintained line representation quality.
2Manufacturing precision
If empirical improvements are made through repeated learning processes to maintain color quality, then color accuracy may improve, but significant time is required
Solution Approach 1:
The patent performs color quality assessment and preprocessing before style transfer to prevent color shifting and maintain color accuracy. By addressing color issues in advance rather than through repeated empirical learning, the system achieves color accuracy without significant time loss.
Solution Approach 2:
The patent implements a feedback mechanism that assesses image quality metrics (including color accuracy) and adjusts preprocessing parameters accordingly. This feedback-driven approach enables the system to achieve color accuracy efficiently without requiring repeated empirical learning processes.
3Manufacturing precision
If image preprocessing is performed to enhance quality before style transfer, then image quality is maintained, but additional processing steps are required
Solution Approach 1:
The patent applies preprocessing operations selectively based on assessed image quality metrics. Rather than applying all preprocessing steps to all images, the system identifies specific quality issues (blur, low contrast, small scale) and applies targeted preprocessing, reducing overall processing complexity while maintaining image quality.
Solution Approach 2:
The patent dynamically adjusts preprocessing parameters based on input image characteristics and quality assessment results. This parameter adaptation allows the system to maintain image quality while minimizing unnecessary processing steps, thereby reducing processing complexity.
Data Source
AI summary
An electronic apparatus for image processing is disclosed. The electronic apparatus includes a memory configured to store a first neural network model trained to transfer a style of an image; and a processor configured to: perform at least one image processing operation on a first image, based on at least one quality of the first image, to obtain an image-processed first image; obtain a second image having a style transferred thereto by inputting the image-processed first image to the first neural network model; based on a color difference between the first image and the second image being greater than or equal to a first threshold value perform a weighted-summation of a brightness value of the first image and a brightness value of the second image to obtain a weighted-summed brightness value, and change the brightness value of the first image to the weighted-summed brightness value; and based on the color difference between the first image and the second image being less than the first threshold value, output the second image.