Machine-Generated Image Recognition Through Multi-Model Contrastive Comparison

The patent segments the recognition task into multiple independent binary classification models, each trained to detect specific characteristics of machine-generated images. Instead of relying on a single comprehensive model, the system divides the detection into multiple specialized classifiers that analyze different aspects (e.g., texture patterns, frequency domains, artifact detection), thereby improving overall recognition accuracy while maintaining system reliability

The patent merges the outputs of multiple binary classification models through ensemble methods. The individual model predictions are combined using techniques such as voting, averaging, or weighted summation to produce a final recognition result. This combination approach leverages the strengths of different models and reduces the impact of individual model errors, thus improving both accuracy and reliability

The patent performs preliminary actions by pre-training multiple binary classification models on diverse machine-generated image data before deployment. The models are prepared in advance with their specific detection specializations established during the training phase. This preliminary preparation allows the system to efficiently process target images during operation without requiring complex real-time coordination between models, thus improving accuracy while managing system complexity

The patent uses copying by generating multiple versions of machine-generated images through different image generation models. These synthesized copies serve as training data for the binary classification models, allowing them to learn various patterns and characteristics of machine-generated content. The copied images enable the models to generalize better without requiring access to the actual target image during training, thereby improving recognition accuracy while keeping the system architecture manageable