Object-Aware Moving Image Compression for Regional Quality

The patent applies different compression rates to different regions of the moving image data based on importance. Specifically, regions containing objects of interest (detected via object detection models) are assigned lower compression rates to maintain high image quality, while other regions are assigned higher compression rates to reduce data amount. This resolves the contradiction by making compression quality non-uniform and adapted to local importance.

The patent segments the moving image data into multiple regions based on object detection results. Each region is then processed independently with its own compression parameters. This segmentation allows the system to preserve quality in important regions while aggressively compressing less important areas, thereby optimizing overall storage capacity without uniformly sacrificing image quality.

The patent identifies objects of interest using object detection models and assigns them protected status. These objects are placed in specific regions that receive lower compression rates, ensuring their image quality is preserved. The system thus reduces overall data amount through compression while maintaining reliability for important objects by applying quality-preserving compression only where needed.

The patent introduces an intermediary layer (the object detection model and region classification system) that analyzes the content of moving image data and determines which regions require quality preservation. This intermediary enables the compression system to intelligently differentiate between important and unimportant regions, applying appropriate compression rates to each, thereby reducing data amount without compromising the reliability of important objects.

The patent performs object detection and region classification before the compression process. By pre-identifying important objects and their regions, the system establishes a quality importance map that guides subsequent compression. This preliminary action simplifies the actual compression processing, as the compression algorithm only needs to apply pre-determined compression rates to pre-identified regions, rather than making complex real-time decisions during compression.

The patent segments the image into regions of different importance levels based on object detection results. This segmentation creates a simplified structure where each region has a designated compression rate, making the compression processing more manageable and less complex than attempting to optimize each pixel individually. The segmented approach allows for efficient implementation using standard compression algorithms with region-specific parameters.