Coefficient rate and significance maps in transform coding

A significance map is a representation of positions of significant transform coefficients. In adaptive coding schemes, this map changes with the sample functions and a large fraction of code bits are used to encode it in low bit rate situations. In this paper, using the component variance as significance measure, a coding scheme is designed to reduce the number of code bits used for the significance map. Although it is a suboptimal approach, the performance is close to the rate distortion limit at a low bit rate. The result also justifies some earlier adaptive schemes using spectral entropy as an indicator of the number of coefficients to be coded.