Most wavelet-based image compression algorithms include a bit plane encoder to transmit all wavelet-transformed coefficients. We define the length of a coefficient as the length of its binary representation. For each bit position l, from the largest length to 1, a significance pass states where the length-l coefficients are located; then a sign pass determines the sign of the length-l coefficients; finally, a refinement pass determines the bit located at position l for all coefficients of length strictly larger than l. Existing algorithms do not use specific methods to encode sign and refinement bits, rather than some variants of arithmetic codification.