Identification of DCT signs for sub-block coding

Abstract

Identification of DCT signs for image coding is presented. In the JPEG DCT coding method the sign of the DCT coefficient is included in the coded signal. Usually 0.1/spl sim/0.2 bit/pixel for the average bit rate is taken for coding these signs. This paper shows these DCT signs can be estimated at the receiver side almost perfectly on the basis of the smoothness cost function, so it is not needed to send this information. The smoothness cost function is the measure of a smooth connection of the sub-block borders. Here as a cost function the absolute squared Laplacian sum is used. And the technique to determine the signs is a genetic algorithm. It gives us an effective solution quickly. The simulation results show that the reconstructed images have adequate picture quality.