【PPT】Image Coding and Data Compression.pdfVIP

Image Coding and Data Compression • Biomedical Images are of high spatial resolution and fine gray-scale quantisiation – Digital mammograms: 4,096x4,096 pixels with 12bit/pixel 32MB per image – Volume data (CT MRI): 512x512x64 with 16bit/pixel 32MB per examination • Health-care jurisdictions: store images ~ 7 years (or whole childhood) 19/04/2005 Coding Compression - Schleimer Veisterä 1 Why digital images? • Films deteriorate • Easier accessible in a database • Multiple copies without expenses • Digital Image Analysis • Allows for compression via coding 19/04/2005 Coding Compression - Schleimer Veisterä 2 Why can we compress images? • Redundancy – Code redundancy – Spatial redundancy (adjacent pixel intensities are correlated) – Psychovisual redundancy (this might not be an issue if one does digital image processing) • Lossless vs. Lossy Compression • “Diagnostically lossless” 19/04/2005 Coding Compression - Schleimer Veisterä 3 Shannon’s Source Coding Theorem • The best achievable lossless compression is bounded by the Entropy. H [x]=l(x)H[x]+1 • Necessary to coding sequences of symbols • You can not do better except with lossy compression • Entropy H [x] is usually unknown. 19/04/2005 Coding Compression - Schleimer Veisterä 4 Fundamental technical terms • Alphabet : set of symbols, e.g.: {0,1}. • Word: finite sequence of symbols. • Code: mapping of words form sources alphabet to code alphabet. • Uniquely decodable: (bijective) code words are uniquely recognisable without separator (blanks). • Instantaneously decodable : no codeword is prefix of an other. • Optimal: if it is inst. decodable and has minimum code length given a