Image Data Compression via Boundary Pixel Extraction

The patent extracts only the boundary pixels between different semantic segments and stores only these boundary pixels in the compressed image file, rather than storing all pixels with their segment labels. This extraction of essential boundary information maintains segmentation capability while dramatically reducing data quantity.

The patent segments the image into distinct semantic portions (e.g., sky, ground, objects) and processes each segment separately by identifying boundary pixels. This segmentation allows for efficient compression by focusing only on the boundaries between segments rather than encoding all pixels uniformly.

The patent extracts only the boundary pixels that define segment boundaries and stores only these critical pixels, eliminating the need to store label information for all interior pixels. This maintains segmentation accuracy at boundaries while dramatically reducing storage requirements for the majority of pixels.

Instead of labeling all pixels (excessive action), the patent applies labels only to boundary pixels (partial action). This partial labeling approach is sufficient to maintain segmentation accuracy since boundaries define semantic regions, while avoiding the storage overhead of labeling every pixel.

The patent segments the image into semantic portions and identifies boundary pixels between these segments. By preserving boundary pixels and their spatial relationships, the compression method maintains essential structural and semantic information while reducing file size through selective storage of only boundary data.

The patent uses boundary pixels as intermediary elements that carry information about semantic segmentations. These boundary pixels act as mediators that encode segmentation information implicitly through their positions and relationships, allowing reconstruction of segment boundaries without storing explicit labels for all pixels.