Image Clustering and Annotation Propagation for Neural Network Training

The patent applies preliminary action by pre-clustering images using histogram-based similarity scores before the actual training process. This pre-processing step groups similar images together, allowing annotations to be propagated efficiently across clusters. By performing this clustering action beforehand, the system reduces the time required for data preparation during the main training workflow, as annotations can be automatically transferred within clusters rather than manually created for every image.

The patent employs copying by propagating annotations from representative images within each cluster to other similar images in the same cluster. Instead of manually annotating every image, the system creates copies of annotations across semantically similar images based on their histogram similarity. This copying approach dramatically reduces annotation time while maintaining consistency across the dataset, directly addressing the time loss issue without sacrificing processing efficiency.

The patent applies segmentation by dividing the image analysis process into distinct functional modules: histogram computation for similarity assessment, clustering for group formation, annotation propagation for labeling, and edge detection for angle estimation. Each module handles a specific aspect of the analysis, making the overall system more manageable and less complex while maintaining comprehensive analysis capabilities. This modular segmentation allows each component to be optimized independently for precision without proportionally increasing overall system complexity.

The patent implements universality by designing a multi-functional processing system that performs multiple tasks within a unified framework. The same clustering mechanism based on histogram similarity serves both for grouping images and for guiding annotation propagation. Additionally, the edge detection module serves dual purposes in both object boundary identification and angle estimation. This multi-functionality reduces device complexity by avoiding redundant specialized components while maintaining high measurement precision across different analysis tasks.