Super-Resolution Image Reconstruction via Dictionary Matching

The patent introduces a dictionary as an intermediary component that stores pre-computed high-resolution image patches. Instead of directly interpolating low-resolution pixels, the system matches low-resolution patches against the dictionary to find corresponding high-resolution patches, thereby mediating the reconstruction process to achieve both speed and quality.

The patent performs preliminary action by pre-computing and storing high-resolution image patches in the dictionary before the actual reconstruction task. This pre-processing allows the reconstruction phase to simply retrieve and match patches, achieving fast reconstruction without sacrificing quality.

The patent merges two approaches: edge-based super-resolution for improving edge quality and dictionary learning for recovering high-frequency texture details. By combining the strengths of both methods, the system achieves comprehensive image restoration that preserves both edges and fine textures.

The patent creates a composite reconstruction approach by integrating edge detection algorithms with dictionary-based patch matching. This composite method leverages the complementary strengths of edge-based techniques (for structural integrity) and dictionary learning (for texture recovery).

The patent replaces the traditional mechanical matching process with a learned similarity metric. Instead of using simple pixel-wise comparison, the system employs a learned similarity measure that captures more nuanced relationships between image patches, thereby improving matching accuracy.

The patent changes the parameter used for matching from simple pixel intensity comparison to a more sophisticated similarity metric that considers local image structures and textures. This parameter change enables more accurate matching of high-frequency details.