Image restoration methods based on quantum field models

The multispectral image restoration method presented is based on an analogy between an image with N pixels and a 2-D solid with N particles on a regular grid. Each pixel is associated with a single particle (SP). The movement of each particle is described using SP models from quantum mechanics. The SP analogy is created by associating the observed feature vectors of each pixel and the noise covariance in the multispectral image with the potential energy function and the total energy of the physical system, respectively. The approach is used to design data-dependent image restoration filters applicable to multispectral images. The filter weights are iteratively updated using a deterministic relaxation scheme to reach a global equilibrium state. The restoration of simulated gray-scale images led to significantly better segmentations than typical spatial image enhancement procedures reported in the literature. The restoration of one real multispectral magnetic resonance image gave a substantial reduction in the noise while enhancing all real structures.<>