A novel context model for remote sensing image compression

Due to the insufficient employment of the correlation among wavelet coefficients, existing significance coding methods can't reduce entropy redundancy efficiently. In order to solve this problem, a significance context model based on intraband and interband correlation is proposed. The model uses neighbor coefficients in the same subband and a parent coefficient in the lower subband as context to predict the current coding coefficient. Neighbor weight and parent weight are defined to distinguish prediction effect of neighbor coefficients and parent coefficient. For neighbor coefficients, different neighbor weight values are assigned according to their directions and bit-planes. Parent coefficient as a significant coefficient has the same prediction effect on either the current bit-plane or above bit-plane, so it is assigned only one weight value. With classifying the coding coefficients according to neighbor weight and parent weight, and merging the contexts with similar probability distribution, the final context classification scheme fitting for most remote sensing images is acquired. Experimental results have shown that the proposed significance context model is prior to the JPEG2000's. It can employ correlation among wavelet coefficients more sufficiently, and remarkably improve the compression performance.