Synthetic aperture radar processing by a multiple scale neural system for boundary and surface representation

Abstract

A neural network model of boundary segmentation and surface representation is developed to process images containing range data gathered by a s y n thetic aperture radar (SAR) sensor. The boundary and surface processing are accomplished by an improved Boundary Contour System (BCS) and Feature Contour System (FCS), respectively, t h a t h a ve been derived from analyses of perceptual and neurobiological data. BCS/FCS processing makes structures such as motor vehicles, roads, and buildings more salient and interpretable to human observers than they are in the original imagery. Early processing by ON cells and OFF cells emb e d d e d i n s h unting centersurround network models preprocessing by lateral geniculate nucleus (LGN). Such preprocessing compensates for illumination gradients, normalizes input dynamic range, and extracts local ratio contrasts. ON cell and OFF cell outputs are combined in the BCS to de ne oriented lters that model cortical simple cells. Pooling ON and OFF outputs at simple cells overcomes complementary processing de ciencies of each c e l l t ype along concave and convex contours, and enhances simple cell sensitivity to image edges. Oriented lter outputs are recti ed and outputs sensitive to opposite contrast polarities are pooled to de ne complex cells. The complex cells output to stages of shortrange spatial competition (or endstopping) and orientational competition among hypercomplex cells. Hypercomplex cells activate long range cooperative bipole cells that begin to group image boundaries. Nonlinear feedback b e t ween bipole cells and hypercomplex cells segments image regions by cooperatively completing and regularizing the most favored boundaries while suppressing image noise and weaker boundary groupings. Boundary segmentation is performed by three copies of the BCS at small, medium, and large lter scales, whose subsequent i n teraction distances covary with the size of the lter. Filling-in of multiple surface representations occurs within the FCS at each scale via a boundary-gated di usion process. Di usion is activated by the normalized LGN ON and OFF outputs within ON and OFF lling-in domains. Di usion is restricted to the regions de ned by gating signals from the corresponding BCS boundary segmentation. The lled-in opponent O N and OFF signals are subtracted to form double opponent surface representations. These surface representations are shown by a n y of three methods to be sensitive to both image ratio contrasts and background luminance. The three scales of surface representation are then added to yield a nal multiple-scale output. The BCS and FCS are shown to perform favorably in comparison to several other techniques for speckle removal.

A neural network for enhancing boundaries and surfaces in synthetic aperture radar images

“…DISCUSSION The major improvements of the present algorithm over that of Grossberg et al (1995) can best. be appreciated by considering Stage 3 of the earlier algorithm, which includes complex cell output at 12 orientations, computed at three scales.…”

“…Our approach further develops the use of surface contrast enhancement, normalization, and diffusive filling-in between form-sensitive boundary signals as a technique for SAR image enhancement, as described by Grossberg et al (1995). Our algorithm capitalizes on the structure-sensitive operations of a neural model of early visual processing in order to enhance the appearance of scenic surfaces based on information distributed across large regions of the image.…”

“…Like the Grossberg et al (1995) algorithm, the present one can be run independently at multiple scales, meaning that boundary signals are detected and completed, and surface representations filled-in, at small, medium and large scales, before a final composite is achieved by averaging. Compared with the Grossberg et a.l.…”

A neuromorphic model for achromatic and chromatic surface representation of natural images

“…For simplicity, three scales are used in the simulation: Small-scale and medium-scale contrast signals and large-scale luminance signals. For simplicity, the size of the sharp center kernel is fixed throughout scales (Grossberg, Mingolla & Williamson, 1995;Mingolla, Ross & Grossberg, 1999;Rahman et al, 1996Rahman et al, , 1997. Figure 3 illustrates how the three different scales respond to luminance inputs.…”

Probabilistic synaptic weighting in a reconfigurable network of VLSI integrate-and-fire neurons

“…One example of a network architecture that lends itself to ef®cient implementation by our scheme is the boundary contour system (BCS) for image segmentation and boundary completion in the presence of clutter and occlusion (Grossberg & Mingolla, 1985;Grossberg, Mingolla & Williamson, 1995;Mingolla, Ross & Grossberg, 1999).…”