A new predictive vector quantization (PVQ) technique capable of explor
ing the nonlinear dependencies in addition to the linear dependencies
that exist between adjacent blocks (vectors) of pixels is introduced.
The two components of the PVQ scheme, the vector predictor and the vec
tor quantizer, are implemented by two different classes of neural netw
orks. A multilayer perceptron is used for the predictive component and
Kohonen self-organizing feature maps are used to design the codebook
for the vector quantizer. The multilayer perceptron uses the nonlinear
ity condition associated with its processing units to perform a nonlin
ear vector prediction. The second component of the PVQ scheme vector q
uantizes the residual vector that is formed by subtracting the output
of the perceptron from the original input vector. The joint-optimizati
on task of designing the two components of the PVQ scheme is also achi
eved. Simulation results are presented for still images with high visu
al quality.