Image Content Moderation Using Fusion-Based Tagging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image content moderation technologies face challenges in accurately categorizing and tagging objects in images, particularly on e-commerce websites, due to the complexity of identifying inappropriate content and assigning correct gender or attributes like size and color, especially when products are displayed by individuals whose gender differs from the product's intended gender.
Innovation Solution
The proposed solution involves an image content moderation apparatus and method that uses a combination of learning models, including convolutional neural networks, to classify objects into categories and tag attributes by analyzing images for inappropriate content, gender, and detailed product information, employing inappropriate content taggers, object category classifiers, and fusion-based taggers to determine appropriate or inappropriate classifications and tags.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional image content moderation methods are used, then the system is simple to implement, but the accuracy of categorizing and tagging objects is insufficient
Solution Approach 1:
The patent segments the image content moderation task into multiple specialized components: inappropriate content detection, object detection, gender classification, and attribute tagging. Each component is handled by a dedicated module that processes specific aspects of the image, allowing for high precision in each sub-task while managing overall system complexity through modular architecture.
Solution Approach 2:
The patent introduces intermediary components such as the inappropriate content tagger that acts as a mediator between the raw image input and the final categorization output. This intermediary layer filters and prepares data before it reaches the main classification systems, improving overall accuracy while maintaining clear system boundaries.
2Productivity
If manual content moderation is used, then the system is easy to operate, but the productivity is low
Solution Approach 1:
The patent implements self-service automation where the system automatically detects, categorizes, and tags image content without human intervention. The inappropriate content tagger and object classification systems operate autonomously, analyzing images and assigning categories based on learned patterns, thereby dramatically increasing productivity while the modular design keeps complexity manageable.
Solution Approach 2:
The patent replaces manual mechanical moderation processes with automated computational systems. Machine learning models and deep learning networks substitute for human reviewers, performing image analysis, object detection, and categorization tasks that were previously done manually, thus enhancing productivity while the systematic approach controls complexity.
3Reliability
If simple tagging methods are used, then the system is easy to manufacture, but the reliability of inappropriate content identification is insufficient
Solution Approach 1:
The patent divides the reliability-critical inappropriate content identification task into separate specialized modules. The inappropriate content tagger focuses exclusively on detecting inappropriate material, while other modules handle benign content categorization. This segmentation allows each module to be optimized for its specific function, improving reliability without requiring the entire system to be overly complex.
Solution Approach 2:
The patent introduces an intermediary inappropriate content tagger that acts as a specialized filter between image input and final output. This mediator layer provides an additional layer of verification and specialized analysis for potentially inappropriate content, enhancing reliability while maintaining clear separation of concerns that manages system complexity.
Data Source
AI summary
In some examples, image content moderation may include classifying, based on a learning model, an object displayed in an image into a category. Further, image content moderation may include detecting, based on another learning model, the object, refining the detected object based on a label, and determining, based on the another learning model, a category for the refined detected object. Further, image content moderation may include identifying, based on the label, a keyword associated with the object, and determining, based on the identified keyword, a category for the object. Further, image content moderation may include categorizing, based on a set of rules, the object into a category, and moderating image content by categorizing, based on aforementioned analysis the object into a category. Yet further, image content moderation may include tagging, based on fusion-based tagging, the object with a category and a color associated with the object.