|
|
|
“Interframe Coding of Digital Images Using Transform and Hybrid Transform/Predictive Techniques”
by John Alan Roese
June 1976
In the design of digital image coding systems, the principal objective is to achieve high quality receiver image reconstructions with a minimum number of transmitted code bits. Bit rate reductions are achieved by exploiting statistical redundancies within an image. This is combined by transmission of only those portions of the mathematical image representation which the human observer is most sensitive to. This dissertation describes research intended to extend current image coding techniques to the coding of sequences of digital images transmitted over a digital communications channel. The emphasis is directed towards definition of an image coding system that exploits temporal as well as spatial image redundancies.
A primary objective of this investigation is to develop a class of interframe hybrid transform/predictive coders having near optimum levels of performance. The interframe hybrid coder implementations considered employ two-dimensional unitary transforms in the spatial domain coupled with first-order DPCM predictive coding in the temporal domain. Based on a statistical image representation, a model is developed for the hybrid coder transform coefficient temporal difference variance matrix. With this model, theoretical MSE performance levels for the hybrid coder with zonal coding are determined as a function of spatial subblock size.
Implementations of the interframe hybrid coder using discrete cosine and Fourier transforms are experimentally evaluated. High quality image reconstruction are demonstrated for reductions of 32:1 in average pixel bit rate. Operational considerations investigated for the hybrid interframe coder include initial conditions, spatial and temporal adaptation, reinitialization, and total transmission bit rate. Also, sensitivity of the interframe hybrid coder to channel error is studied.
Comparisons are drawn between hybrid transform/predictive and three-dimensional transform interframe coders. Theoretical zonal sampling and zonal coding MSE performance for three-dimensional cosine transform coders are evaluated for different frame storage requirements and spatial subblock sizes.
A tabular summary of experimental performance and system design parameters for the main classes of interframe coders is presented.