Contour-Detail Image Processing for Controlled Style Generation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning models, such as diffusion models, can generate unintended images due to uncontrolled parameter adjustments, leading to unsatisfactory output results.
Innovation Solution
An image processing apparatus that preprocesses input images by separating contour and detail components, generating a composite image, and then inputs this composite to a machine learning model to control the output style, using techniques like LoRA for parameter adjustment.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If machine learning models are used to generate images with adjusted parameters, then image generation flexibility is improved, but image accuracy and reliability deteriorate
Solution Approach 1:
The input image is segmented into multiple components: contour image, detail image, and style image. This segmentation allows each component to be processed independently and combined later, ensuring that the structural accuracy from contour images is preserved while incorporating stylistic transformations through the machine learning model.
Solution Approach 2:
The contour and detail images are prepared in advance before being input to the machine learning model. By pre-processing these images to extract essential structural and textural information, the system ensures that the generated output maintains accurate representation of the original image's contours and features while allowing stylistic flexibility.
2Device complexity
If machine learning models generate images directly from input images, then processing simplicity is improved, but output control and precision deteriorate
Solution Approach 1:
The image processing is divided into distinct stages: extracting contour information, extracting detail information, generating style image, and combining results. This segmentation provides precise control over each aspect of image generation while maintaining a relatively simple overall workflow through automated processing at each stage.
Solution Approach 2:
The contour image and detail image serve as intermediaries between the original input image and the final generated output. These intermediate representations allow the system to maintain control over structural accuracy while enabling flexible stylistic transformation through the machine learning model.
3Adaptability or versatility
If parameter adjustment techniques like LoRA are used, then style adaptation capability is improved, but system complexity and computational requirements worsen
Solution Approach 1:
The LoRA technique applies low-rank decomposition to specific weight matrices in the machine learning model, adjusting only the necessary parameters for style adaptation rather than the entire model. This localized parameter adjustment maintains style adaptation capability while reducing overall system complexity and computational requirements.
Data Source
AI summary
An image processing apparatus comprises a controller. The controller is configured to perform obtaining a target image representing an object, obtaining a contour image representing a contour of the object and a detail image representing more fine features of the object, generating a composite image by composing multiple images including the contour image and the detail image, and obtaining a new image by inputting the composite image to a machine learning model.