Fan-Shaped Image Frame Segmentation for Paper Cup Models

The patent segments the image processing task into multiple stages: rough positioning using geometric features, followed by precise positioning using template matching and feature point matching. This multi-stage segmentation allows each stage to focus on specific aspects, improving overall precision while maintaining stability through the coordinated operation of multiple processing steps.

The patent performs preliminary rough positioning using geometric features before conducting precise template matching. This preliminary action prepares the image data by establishing initial boundaries and reducing the search space, which stabilizes subsequent processing and prevents noise from causing extreme deviations in final results.

The patent introduces an intermediary automated annotation system that bridges manual annotation and final precise annotation. This intermediary system uses pre-trained neural networks to generate initial annotations, which are then refined through automated correction algorithms. This intermediary layer maintains ease of implementation while significantly improving annotation precision, as it combines automated processing with targeted manual verification only where needed.

The patent replaces purely manual mechanical annotation processes with an automated system based on neural networks and algorithmic processing. This substitution maintains simplicity of operation while dramatically improving precision, as the automated system can consistently apply complex geometric and photogrammetric calculations without the variability and errors inherent in manual processes.

The patent implements a feedback mechanism where the initial segmentation mask from the neural network is evaluated, and its boundaries are refined through iterative adjustment. The system compares the initial automatic segmentation with expected geometric constraints and template matching results, then feeds this information back to correct the segmentation boundaries. This feedback loop maintains full automation while significantly improving precision by correcting systematic errors in the initial neural network output.

The patent changes key parameters of the segmentation process by adjusting the neural network output through geometric constraints and template matching. Instead of directly using the raw network output, the system modifies parameters such as boundary positions, angles, and dimensions based on additional processing steps. This parameter transformation maintains automation while improving precision by incorporating multiple sources of information and correcting deviations from expected geometric properties.