Computer-generated optical multiwavelet filters for hybrid image-classification systems.

Optical coherent Fourier correlators are applicable in real-time image analysis such as image classification. The functionality of Fourier correlators can be increased by use of multifunctional filters, which have many spatially multiplexed impulse responses. The concept of multiresolution analysis on the basis of wavelet theory offers profitable methods to design multifunctional filters for image analysis. The applicability of such filters is demonstrated by an example in which different characteristic textures of medical images are extracted. The physical implementation of multiwavelet filters is restricted by modulation-domain constraints imposed by the use of spatial-light-modulator or of diffractive-element fabrication technology. Coding methods of diffractive optics are shown to be helpful to transform the original complex-valued distributions of multiwavelet filters into light-efficient quantized phase-only distributions by preservation of the original filter functionality. The quality of the designed diffractive phase filters is documented by computer experiments.