Inference optical filters for distortion-invariant scene analysis
暂无分享,去创建一个
To handle multiple objects in several classes in clutter with distortions present, we use a hierarchical inference optical correlator using sets of new filters [1]. We first employ morphological processing operations and wavelet filters to reduce clutter. The morphological operations are nonlinear image processing functions. The wavelet filters use multiresolution methods. Figure la shows an input scene with 6 targets in severe clutter. The morphological operations used [2] (closure minus opening) remove large area clutter regions of different intensity and convert the polarity of all targets to bright (Figure lb). The wavelet processing [2] locates regions of clutter particles smaller than the targets (Figure 1c). Removing these from the morphological result in Figure lb yields an input with greatly reduced clutter (Figure 1d). This completes our first level of filters for clutter reduction.
[1] David Casasent. An optical correlator feature extractor neural net system , 1992 .
[2] David Casasent,et al. Optical morphological processors: gray scale with binary structuring elements, detection, and clutter reduction , 1992, Other Conferences.