A construction of pattern classification system using direct‐drawn computer‐generated fourier hologram

This paper considers the construction of the trainable classification system by redrawing a computer-generated Fourier hologram (CGFH), which is based on the previously proposed method of pattern classification using CGFH. CGFH has a flexibility that can be synthesized by computation, and it is expected that a flexible pattern classifier will be constructed if the adjustment of the weight in the proposed system can be realized through the redrawing of CGFH. However, if CGFH is constructed by the traditional method by the size-reduction of a photograph, the required processing such as size reduction and developing will make it difficult to redraw CGFH on the optical system. To solve this problem, this paper considers a system which generates CGFH directly on the glass substrate, by a laser beam scanning on a thin film. Using the proposed direct-drawn CGFH (DDCGFH) system, a basic experiment is made on the construction of the pattern classification system by learning (i.e., redrawing of CGFH). First, an investigation is presented on the coding of CGFH suited to the DDCGFH construction system, together with a result of experiment for DDCGFH. Then the construction of the classification system by redrawing DDCGFH is considered. The algorithm and the result of basic experiment are reported.