Abstract: An image processing system includes an input part for reading an image from a color document containing a multilevel gradation graphic image and a text image, a region separator part for separating the image into a plurality of elemental blocks having M.times.N picture elements according to a type of image data that is contained in each elemental block, whether the elemental block contains a black-and-white line portion of the image or a halftone portion of the image, and an encoding part for encoding image data in the black-and-white line portion through an encoding process appropriate for bi-level image data and for encoding image data in the halftone portion through an encoding process appropriate for multilevel gradation image data so as to produce a compressed image.

Abstract: A voice signal coding method for compressing voice data by coding is provided. PCM coded voice data are grouped into blocks in a timed sequence, each block having a predetermined number of data. A scale data representing a MSB corresponding to a maximum value in absolute value in each block is identified. Then, a code data is defined by a predetermined number of bits of data containing the MSB, thereby having the voice data compressed by coding. The code data is corrected such that an error between a decoded value of the code data and the original voice data corresponding to the code data is minimized. With such a correction step, the code data is refined such that a reproduced voice signal by decoding the thus corrected code data is very close to the original voice signal.