Arithmetic coding for image compression with adaptive weight-context classification

• We introduce a new binary arithmetic coding strategy for bitplane coding.
• Classifying the weight context can reduce the context-dilution degree.
• The result of a context model with 14 neighbors is better than that of a context model of eight neighbors.
• On the fly strategy can achieve higher compression performance compared with off-line strategy.

In this paper, a new binary arithmetic coding strategy with adaptive-weight context classification is introduced to solve the context dilution and context quantization problems for bitplane coding. In our method, the weight, obtained using a regressive–prediction algorithm, represents the degree of importance of the current coefficient/block in the wavelet transform domain. Regarding the weights as contexts, the coder reduces the context number by classifying the weights using the Lloyd–Max algorithm, such that high-order is approximated as low-order context arithmetic coding. The experimental results show that our method effectively improves the arithmetic coding performance and outperforms the compression performances of SPECK, SPIHT and JPEG2000.