| 摘要 |
The restoration of images by vector quantization utilizing visual patterns is disclosed. One disclosed embodiment comprises restoring detail in a transition region of an unrestored image, by first identifying the transition region and forming blurred visual pattern blocks. These blurred visual pattern blocks are compared to a pre-trained codebook, and a corresponding high-quality visual pattern blocks is obtained. The high-quality visual pattern block is then blended with the unrestored image to form a restored image. |
