Abstract | ||
---|---|---|
Recently, an example-based super-resolution (SR) method specifically for face image, which takes the correspondence of facial parts into consideration, was proposed. The method constructs a database of face-image patches with their positions from example images of normalized faces. Given a low-resolution (LR) image, the method generates an SR image by finding similar patches in the database and incorporating high-frequency component of the patches to the LR image. Since the method just divides an example image into patches, for example, 5x5 patches in five-pixels intervals, the SR process may misses the chance to have the patch of the finest position. The proposed method generates patches in one-pixel intervals to increase the chance. In order to cope with the massively increased size of the patch database, the proposed method restricts the area to search the patch candidates for each position. An experimental result shows the proposed method achieves better results than the previous in term of the peak signal-to-noise ratio (PSNR). Furthermore, the proposed method has reduced computational time by 90% compared to the previous.
|
Year | DOI | Venue |
---|---|---|
2019 | 10.1145/3309074.3309109 | Proceedings of the 3rd International Conference on Cryptography, Security and Privacy |
Keywords | Field | DocType |
computational time, example-based, face image, facial parts, patch-searching area, super-resolution | Normalization (statistics),Pattern recognition,Computer science,Artificial intelligence,Superresolution | Conference |
ISBN | Citations | PageRank |
978-1-4503-6618-2 | 0 | 0.34 |
References | Authors | |
0 | 4 |
Name | Order | Citations | PageRank |
---|---|---|---|
Suhail Hamdan | 1 | 0 | 0.34 |
Yohei Fukumizu | 2 | 2 | 1.97 |
Tomonori Izumi | 3 | 34 | 20.88 |
Hironori Yamauchi | 4 | 12 | 4.74 |