Visual Communication - Information Rate and Robustness of Performance

Information Rate and Robustness of Performance

The informationally-optimized design of visual communications systems is based inherently on some prescribed statistical properties of natural (random) scenes given by the normalized power spectral density (PSD) p-2p with variance p2, as illustrated below, where is the mean spatial detail of the scene relative to the spacing of the sampling intervals of the image-gathering process are the spatial frequency coordinate system with units of cycles/sample, and m controls the fall-off rate with increasing spatial frequencies.

Robustness of Electro-optical Design

The selection of the spatial frequency response SFR of the image-gathering device relative to its sampling passband , as controlled by the optical-response index , for maximum information rate may be based on the assumption that m=3 and =1, as summarized in the table. Similar plots of for various values of m, ranging from 1.4 to 4 lead to the selection of for the available SNR, as would plots of for various values of .


Robustness of Image Restoration

This last figure presents the information rate and the corresponding maximum-realizable fidelity for matched and mismatched Wiener restorations. The matched restorations use the correct values of m and . These curves reveal that the informationally optimized designs produce the

highest fidelity and robustness, and that both improve with increasing . Ordinarily, to assure both a high fidelity and robustness, it is necessary for the SNR to be high, above ~64, so that an information rate 3 bits/sample can be attained. Moreover, these curves reveal that, ordinarily, one cannot go far wrong by assuming that the mean spatial detail is equal to the sampling intervals (i.e., =1). This observation appeals intuitively because spatial detail that is much smaller cannot be resolved, and detail that is larger is not degraded substantially by blurring and aliasing.



Information efficiency

Home Site map

Web site curator: Glenn Woodell