Optimum coding of binary data for holographic matched filters

The searching of large data bases to retrieve unindexed information involves search times of many seconds. It is clearly an operation where a high order of parallelism could pay dividends and hence an optical approach based on a holographic matched filter looks interesting. The matched filter system consists of an input plane, a Fourier transforming lens, a holographic recording or Fourier transform plane (FTP), a second lens giving an inverse Fourier transform and an output plane. Information in the input plane consists of a block of data (the object) which is recorded as a hologram in the FTP in conjunction with a reference beam which is a point source. It is required to search the object to find specific character strings (scenes) or fields within it. Having found the position of the scene, the relevant associated data can be accessed optically or electronically. The `raw´ information is binary data which is represented in a conventional octet form internally in the computer but can be recoded into a form that is optimum for optical processing. The authors discuss how best to code the information to produces adequate discrimination between a string that is a perfect match and one that is a near match