Structural Similarity Index with Predictability of Image Blocks
Tutkimustuotos › › vertaisarvioitu
Yksityiskohdat
Alkuperäiskieli | Englanti |
---|---|
Otsikko | 2018 IEEE 17th International Conference on Mathematical Methods in Electromagnetic Theory, MMET 2018 - Proceedings |
Kustantaja | IEEE COMPUTER SOCIETY PRESS |
Sivut | 115-118 |
Sivumäärä | 4 |
Vuosikerta | 2018-July |
ISBN (painettu) | 9781538654385 |
DOI - pysyväislinkit | |
Tila | Julkaistu - 10 syyskuuta 2018 |
OKM-julkaisutyyppi | A4 Artikkeli konferenssijulkaisussa |
Tapahtuma | IEEE International Conference on Mathematical Methods in Electromagnetic Theory - Kyiv, Ukraina Kesto: 2 heinäkuuta 2018 → 5 heinäkuuta 2018 |
Julkaisusarja
Nimi | |
---|---|
ISSN (elektroninen) | 2161-1750 |
Conference
Conference | IEEE International Conference on Mathematical Methods in Electromagnetic Theory |
---|---|
Maa | Ukraina |
Kaupunki | Kyiv |
Ajanjakso | 2/07/18 → 5/07/18 |
Tiivistelmä
Structural similarity index (SSIM) is a widely used full-reference metric for assessment of visual quality of images and remote sensing data. It is calculated in a block-wise manner and is based on multiplication of three components: similarity of means of image blocks, similarity of contrasts and a correlation factor. In this paper, two modifications of SSIM are proposed. First, a fourth multiplicative component is introduced to SSIM (thus obtaining SSIM4) that describes a similarity of predictability of image blocks. A predictability for a given block is calculated as a minimal value of mean square error between the considered block and the neighboring blocks. Second, a simple scheme for calculating the metrics SSIM and SSIM4 for color images is proposed and optimized. Effectiveness of the proposed modifications is confirmed for the specialized image databases TID2013, LIVE, and FLT. In particular, the Spearman rank order correlation coefficient (SROCC) for the recently introduced FLT Database, calculated between the proposed metric color SSIM4 and mean opinion scores (MOS), has reached the value 0.85 (the best result for all compared metrics) whilst for SSIM it is equal to 0.58.