Context-based filtering for document image compression

Two context-based filtering methods are introduced. The methods are based on statistical context modeling by gathering context-wise statistics from the entire image. Uncommon pixels in low information contexts are unconditionally inverted (simple context filter), or inverted conditionally depending whether the gain in compression overweighs the loss of information (gain-loss filter). Both methods improve compression performance of the document images while preserving the image quality, measured as the OCR accuracy. The gain-loss filter is capable of reaching the compression limit estimated by the compression of the noiseless digital original.