Image Style Classification Using Procedural Style Embeddings
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing media editing software lacks the ability to classify unseen image styles effectively and requires extensive manual user input, leading to inefficient image style transfer that is often destructive and resource-intensive.
Innovation Solution
A system that classifies image styles by determining a similarity score between a target image and predetermined styles, allowing for non-destructive, editable application of image styles through procedural manipulation and reduced manual input.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If image transfer technologies apply image styles to other images using wholesale changes in a single forward pass, then the image style transfer is achieved quickly, but the functionality becomes destructive and users have no control over different layers of the image style transferred
Solution Approach 1:
The patent segments the image style transfer process into multiple independent layers, each representing different aspects of the style (e.g., color, texture, lighting). These layers can be individually controlled and adjusted, allowing users to modify specific aspects without affecting the entire image style transfer.
Solution Approach 2:
The system transitions from a static, all-or-nothing style transfer approach to a dynamic, multi-layered approach where each layer can be independently adjusted. This allows the system to adapt to user preferences by enabling or disabling specific style layers based on what the user wants to preserve or modify.
2Adaptability or versatility
If media editing software provides extensive tools for modifying visual data, then the functionality and control over images are enhanced, but the complexity of the software increases and requires extensive manual user input
Solution Approach 1:
The system automatically performs image style classification and layer identification without requiring extensive manual user input. The machine learning models autonomously analyze the target image, identify the desired style characteristics, and separate the style into controllable layers, allowing the software to serve itself rather than requiring complex manual configuration.
Solution Approach 2:
The patent changes the fundamental parameters of how image style is represented and applied, transitioning from a single monolithic style parameter to multiple independent style layers. This parameter transformation simplifies the user interface by allowing selection through similarity scoring rather than requiring manual adjustment of numerous editing parameters.
3Productivity
If image style transfer technologies apply wholesale changes to images, then the style transfer is achieved in a single pass, but computing resources such as disk I/O are consumed unnecessarily
Solution Approach 1:
The system extracts only the essential style characteristics from the reference image and represents them as procedural parameters rather than applying the entire image data. This extraction approach retrieves only the necessary stylistic information (colors, textures, patterns) and discards redundant data, significantly reducing disk I/O and computing resource consumption.
Solution Approach 2:
The patent performs preliminary analysis of the target and reference images to identify and extract the relevant style parameters before the actual style transfer process. This preliminary action includes classifying the image styles, identifying matching characteristics, and preparing the procedural parameters in advance, which optimizes the subsequent style application and reduces unnecessary resource consumption during the main processing phase.
Data Source
AI summary
Various disclosed embodiments are directed to classify or determining an image style of a target image according to a consumer application based on determining a similarity score between the image style of a target image and one or more other predetermined image styles of the consumer application. Various disclosed embodiments can resolve image style transfer destructiveness functionality by making various layers of predetermined image styles modifiable. Further various embodiments resolve tedious manual user input requirements and reduce computing resource consumption, among other things.