1:00, IMDSP-L6.1
A ROBUST KALMAN FILTER DESIGN FOR IMAGE RESTORATION
Y. CHEE, Y. SOH
In image deconvolution or restoration using Kalman filter, the image and blur models are required to be known for the restoration process. Generally, the accuracy of the restoration depends on the accuracy of the given models. Unfortunately, the image and blur models are normally unknown in practice. To solve the problem, an identification stage is employed to estimate the image and blur models. However, the estimated models are seldom accurate especially with the presence of noise in the image. This paper presents a robust Kalman filter design for image deconvolution that can accommodate the inaccuracy in the estimated image and blur models. If the inaccuracy can be modelled as addictive white Gaussian noise with a known variance, it can be stochastically account for in the robust filter design. In the simulation tests performed, the robust design achieved improved accuracy in the image restoration even though inaccurate image and blur models were used.

1:20, IMDSP-L6.2
EDGE ADAPTIVE RESTORATION OF NOISY, BLURRED IMAGES
G. FOSTER, N. NAMAZI
We present a method to find the mean square estimate of the original image using a Gauss-Markov image model and a known point spread function. The performance of the edge adaptive technique compares favorably to the Wiener filter on Synthetic and real images with mild linear (motion) blur and additive white Gaussian noise.

1:40, IMDSP-L6.3
AN ATTRACTOR SPACE APPROACH TO BLIND IMAGE DECONVOLUTION
K. YAP, L. GUAN
In this paper, we present a new approach to adaptive blind image deconvolution based on computational reinforced learning in attractor-embedded solution space. A new subspace optimization technique is developed to restore the image and identify the blur. Conjugate gradient optimization is employed to provide an adaptive image restoration while a new evolutionary scheme is devised to generate the high-performance blur estimates. The new technique is flexible as it does not suffer from various image or blur constraints imposed by most traditional blind methods. Experimental results show that the new algorithm is effective in blind deconvolution of images degraded under different blur structures and noise levels.

2:00, IMDSP-L6.4
GENERATION OF SUPER-RESOLVED IMAGES FROM BLURRED OBSERVATIONS USING MARKOV RANDOM FIELDS
D. RAJAN, S. CHAUDHURI
This paper presents a new technique for generating a high resolution image from a blurred image sequence; this is also referred to as super-resolution restoration of images. The image sequence consists of decimated, blurred and noisy versions of the high resolution image. The high resolution image is modeled as a Markov random field (MRF) and a maximum aposteriori (MAP) estimation technique is used. A simple gradient descent method is used to optimize the functional. Further, line fields are introduced in the cost function and optimization using Graduated Non-Convexity (GNC) is shown to yield improved results. Lastly, we present results of optimization using Simulated Annealing (SA).

2:20, IMDSP-L6.5
SIMULTANEOUS IMAGE FORMATION AND MOTION BLUR RESTORATION VIA MULTIPLE CAPTURE
X. LIU, A. EL GAMAL
Advances in CMOS image sensors enable fast image capture, which makes it possible to capture multiple images within a normal exposure time. An algorithm that takes advantage of this capability by simultaneously constructing a high dynamic range image and performing motion blur restoration from multiple image captures is described. The algorithm comprises two main procedures -- photocurrent estimation and motion/saturation detection. It operates completely locally -- each pixel's final value is computed using only its captured values, and recursively, requiring the storage of only a constant number of values per pixel independent of the number of images captured. These modest computational and storage requirements make it feasible to integrate all needed memory and processing with the image sensor on a single CMOS chip. Simulation results demonstrate the enhanced SNR, dynamic range, and the motion blur restoration obtained using our algorithm.

2:40, IMDSP-L6.6
UNWRAPPING PHASE IMAGES BY PROPAGATING PROBABILITIES ACROSS GRAPHS
R. KOETTER, B. FREY, N. PETROVIC, D. MUNSON, JR.
Phase images are derived from source images by applying a modulus operation to each pixel value. Phase unwrapping is the problem of inferring the original, unwrapped values from the wrapped values, using prior knowledge about the smoothness of the image. One approach to solving this problem is to infer the gradient vector field of the unwrapped image and then integrate the gradient field. The gradient in a particular direction at a pixel is equal to the observed pixel difference plus an unknown integer number of shifts. We introduce a technique for inferring these shifts using the low-complexity probability propagation algorithm, applied in a graphical model that prefers shifts that match the phase image and that constrains the shifts to satisfy the properties of a gradient field. We present results for a phase image from the region of the Sandia National Laboratories.