A new algorithm for fractal image compression is developed to speed up the
encoder. This new method converts image blocks into frequency-like domains
using the Hadamard transform, in which the computations of the best matched
are performed. At each search entry, the best mean square error computatio
ns of the eight dihedral symmetries are reformulated into the form of inner
products. By a precise derivation, all redundant computations are complete
ly avoided. With this improved technique, the complexity of the encoder is
substantially reduced. A simulation shows that, with the same PSNR and comp
ression ratio, the new method requires less computation time than the basel
ine method.