A radiometric/geometric transform has been developed which generates images with approximately widesense stationary first (mean)- and second (autocorrelation)-order statistics [R. N. Strickland, Appl. Opt. 22, 1462 (1983)]. The transform is found to enhance the performance of predictive coding. The radiometric transform reduces radiometric redundancy in the image data and, therefore, aids efficient quantization at low bit rates. A spatial transformation, or warp, is applied to produce nearly uniform autocorrelation length throughout the data. This affords a convenient means of reducing spatial redundancy by varying the spatial resolution in the transformed image. Final bit rates of ˜0.6 bits/pixel are realized for high-quality images. We discuss hybrid digital/optical implementation of the stationary transforms for data compression.
© 1983 Optical Society of America
Robin N. Strickland and Warren E. Smith, "Stationary transform processing of digital images for, data compression," Appl. Opt. 22, 2161-2168 (1983)