UBC Theses and Dissertations
Rate control in bi-level image coding Tompkins, David Andrew Douglas
As we continue to pursue the illusive goal of a paperless society, we are increasingly digitizing our documents, and despite increases in computer speeds and capacities, the requirement for data compression is still paramount. The JBIG2 standard is the latest international bi-level image compression standard, and the first that supports lossy coding. A method is proposed that can achieve rate control while coding bi-level images with JBIG2. For compound images, the image is segmented into text and non-text regions. Non-text regions are lossy coded as generic regions with a bit-flipping approach. The bit-flipping approach has a compression limit, so for higher compression the region is coded as a halftone. For halftone regions, an appropriate grid size is selected and a reduced multi-level image is constructed. The multi-level image is lossy coded with either a bit-flipping approach or with a vector quantization approach, depending on target rate. For text regions, rate control is achieved by adjusting the number of unique symbols in the image. The symbols are a subset of the complete set in the lossless image, and are chosen to minimize the distortion. While the rate control in non-text regions is optimized in the mean squared error sense, the rate control for text regions can be adjusted for any distortion measure. To complete the rate control for a compound image, a weighting is assigned to each region so that the overall distortion is minimized. Results demonstrate that the proposed method is effective, and can produce a JBIG2 bitstream at any target compression rate for any bi-level image.
Item Citations and Data