In this article, a new prediction method based on
least-square minimization was added to the MMP (Multidimensional Multiscale Parser) algorithm, in order to improve its rate distortion performance.
A local context is used to adaptively adjust the linear prediction
coefficients, that implicitly embed the local texture characteristics.
Since the decoder repeats the same prediction process using
the reconstructed data, no extra overhead is needed for signaling
the coefficients. This new context-adaptive linear prediction mode was added to the existing intra prediction modes, and the best mode is chosen through rate-distortion optimization.
The experimental results presented in this paper show that
the new prediction mode is able to increase considerably MMPs
coding performance for smooth images, when compared with
state-of-the-art, transform-based methods, and no performance
loss was detected, when applied to non-smooth images.