Title :
Robust classification of blurred imagery
Author :
Kundur, Deepa ; Hatzinakos, Dimitrios ; Leung, Henry
Author_Institution :
Dept. of Electr. & Comput. Eng., Toronto Univ., Ont., Canada
fDate :
2/1/2000 12:00:00 AM
Abstract :
We present two novel approaches for the classification of blurry images. It is assumed that the blur is linear and space invariant, but that the exact blurring function is unknown. The proposed fusion-based approaches attempt to perform the simultaneous tasks of blind image restoration and classification. We call such a problem blind image fusion. The techniques are implemented using the nonnegativity and support constraints recursive inverse filtering (NAS-RIF) algorithm for blind image restoration and the Markov random field (RIRF)-based fusion method for classification by Schistad-Solberg et al. (see IEEE Trans. Geosci. Remote Sensing, vol.32, p.768-78, 1994). Simulation results on synthetic and real photographic data demonstrate the potential of the approaches. The algorithms are compared with one another and to situations in which blind blur removal is not attempted
Keywords :
Markov processes; filtering theory; image classification; image restoration; inverse problems; random processes; recursive filters; Markov random field based fusion method; NAS-RIF algorithm; blind image classification; blind image fusion; blind image restoration; blurred imagery; blurry image classification; exact blurring function; fusion-based approach; linear invariant blur; nonnegativity recursive inverse filtering; real photographic data; robust classification; simulation results; space invariant blur; support constraints recursive inverse filtering; synthetic photographic data; Degradation; Filtering algorithms; Fuses; Helium; Image classification; Image fusion; Image restoration; Markov random fields; Multispectral imaging; Robustness;
Journal_Title :
Image Processing, IEEE Transactions on
