Blind identification of multichannel FIR blurs and perfect image restoration

Authors:

G. B. Giannakis and Robert W. Heath Jr

Reference:

Proc. IEEE Int. Conf. on Image Processing, vol. I., pp. 713-720, Lausanne, Switzerland, September 16-19, 1996.

Abstract:

Despite its practical importance in image processing and computer vision, blind blur identification and restoration has so far been addressed under restrictive assumptions such as all-pole stationary image models blurred by zero-or minimum-phase PSFs. Relying upon diversity (sufficient number of multiple blurred images), we develop blind FIR blur identification and order determination schemes. Apart from a minimal persistence of excitation condition (also present with non-blind setups), the inaccessible input is allowed to be deterministic or random and of unknown color or distribution. With the blurs also satisfying a certain co-primeness condition, we establish existence and uniqueness results which guarantee that single-input/multiple-output FIR blurred images can be restored perfectly but blindly using linear FIR filters. The resulting direct blind restoration filters by-pass the blur identification step and work well in higher SNRs. In lower SNRs better results are achieved by first identifying the blurs and then constructing the MMSE solution.

Keywords:

Blind deconvolution, blind image restoration, perfect reconstruction

This paper is available on IEEE Xplore file