Real-Time Motion Blur Estimation and Restoration in Foveal Active Vision Systems

Foveated, log-polar, or space-variant image architectures provide a high resolution and wide field workspace, while providing a small pixel computation load. These characteristics are ideal for mobile robotic and active vision applications. A common problem in these application areas is image blur and motion artifact. Recently, there has been described a generalization of the Fourier Transform (the fast exponential chirp transform) which allows frame-rate computation of full-field 2D frequency transforms on a log-polar image format. In the present work, we use Wiener filtering, performed using the Exponential Chirp Transform, on log-polar (foveated) image formats to de-blur images which have been degraded by camera motion or other sources of blur. Since the exponential chirp transform provides size and rotation invariant properties, we only need to pre-compute and store a single motion vector template, exploiting the size and rotation properties of the chirp transform. This provides real-time performance without requiring a large storage of pre-computed templates. Examples of this processing on different samples from a data base of images shows effective image restoration, with frame-rate performance on modest hardware (e.g. a Pentium or SHARC DSP processor). This work shows that the fast exponential chirp transform provides a simple algorithm with low spatial complexity for full frame, real-time image restoration of foveal imagery