Image Harmonization via Context Feature Extraction and Semantic Reconstruction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image harmonization technologies struggle to effectively harmonize target images with significantly different colors and textures, resulting in poor harmonization effects, especially when inserting advertisements into videos with vastly different content.
Innovation Solution
An image processing method and apparatus that perform context feature extraction and multi-level semantic information extraction on harmonized images, followed by image reconstruction based on this information to improve the harmonization effect.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If current image harmonization technology is used to harmonize target images with significantly different colors and textures, then the processing speed is maintained, but the harmonization effect deteriorates
Solution Approach 1:
The patent segments the harmonization process into multiple stages: obtaining source and target images, performing image harmonization, and conducting post-processing adjustments. It also segments the target image into different regions (foreground and background) and applies different harmonization strategies to each region, allowing adaptive handling of various content types with significantly different colors and textures.
Solution Approach 2:
The patent implements dynamic adjustment of harmonization parameters based on the content characteristics of the source and target images. The system automatically adjusts color mapping, brightness, and contrast parameters according to the specific content being harmonized, enabling effective harmonization across diverse content types while maintaining processing efficiency.
2Reliability
If traditional image harmonization methods are applied, then the process is simple and fast, but the color and texture consistency between source and target images deteriorates
Solution Approach 1:
The patent introduces an intermediary color mapping module that establishes a mapping relationship between the color spaces of source and target images. This intermediary mechanism transforms colors from the target image to match the color characteristics of the source image, achieving superior color and texture consistency while maintaining a relatively simple processing framework.
Solution Approach 2:
The patent dynamically adjusts multiple parameters including color mapping parameters, brightness adjustment parameters, and contrast parameters based on the statistical characteristics of the source and target images. By changing these parameters adaptively, the system achieves high color and texture consistency without requiring complex processing algorithms.
3Reliability
If basic image harmonization is performed, then the processing time is short, but the overall image quality and naturalness deteriorate
Solution Approach 1:
The patent performs preliminary analysis of the source and target images to extract key characteristics such as color distribution, brightness levels, and contrast ratios before executing the harmonization process. This preliminary action allows the system to pre-calculate optimal harmonization parameters, thereby achieving high image quality and naturalness without significantly increasing processing time.
Solution Approach 2:
The patent applies different harmonization strategies to different regions of the target image. The foreground region containing the advertisement receives different treatment compared to the background region, with localized adjustments to color, brightness, and contrast. This local quality approach enhances overall image quality and naturalness while maintaining efficient processing.
Data Source
AI summary
Embodiments of the present disclosure disclose an image processing method and apparatus. The method may include obtaining a harmonized image. The harmonized image may be harmonized with a promotional content image. The method may further include performing context feature extraction on the harmonized image to obtain context feature information of the harmonized image and extracting multi-level semantic information of an object in the harmonized image based on the context feature information. The method may further include performing image reconstruction based on the context feature information and the multi-level semantic information to obtain a reconstructed image. This solution can improve an image harmonization effect.