Page 131 - Digital Analysis of Remotely Sensed Imagery
P. 131

102    Cha pte r  T h ree

                   Stored in the compressed image are only the transform coefficients
               rather than pixel values of the image. During uncompression the
               JPEG decoder re-creates the normalized transform coefficients first.
               This can be achieved easily via a lookup table as they are coded
               using the uniquely decodable Huffman coding. These coefficients are
               plugged into the formula to generate pixel values that best represent
               the original ones.  An approximate version of the original 8    8
               subarea is created via the inverse transform. Tiling of all subimages
               restores the original uncompressed image. JPEG compression has the
               drawback of introducing artifacts along the border of subimages after
               they are mosaicked to form the original image, because each subimage
               is compressed separately. This problem disappears with JPEG 2000.
                   JPEG 2000 is a new image compression standard that is backward
               compatible with, and further extends, the current standard JPEG with
               increased flexibility in image compression and in access to the
               compressed data. Unlike JPEG that can compress RGB imagery at
               most, JPEG 2000 is able to compress images up to 256 bands. A very
               large compression ratio can be achieved with very little appreciable
               degradation in image quality. Unlike JPEG, which uses a Fourier
               transform, JPEG 2000 uses two coding modes, DCT coding and
               wavelet coding, to achieve more efficient compression.
                   Based on wavelet technology, JPEG 2000 enables an image to be
               compressed with or without loss of information. Error-free
               compression is achieved using a biorthogonal, 5/3 coefficient scaling
               and wavelet vector at an expected compression ratio of 2:1 (Le Gall
               and Tabatabai, 1988). Ordinary lossy compression, if implemented
               with a 9/7 coefficient scaling-wavelet vector, is able to achieve a
               compression ratio of up to 200:1 (Antonini et al., 1992). Such a large
               ratio is achieved because the “mother wave,” which best represents
               the wavelet signature generated from scanning an image, does not
               accompany the compressed image data. Instead, the JPEG 2000
               decoder is equipped with a universal mother wave. Whenever the
               decoder is supplied with a compressed image, it can detect the mother
               wave used. Furthermore, lossy compression can be embedded into
               lossless compression. Since an image can be regarded as composed of
               different regions of interest (e.g., images embedded into text), different
               compression schemes can be applied to different regions of interest in
               the image to help preserve the image quality of those regions. The
               shape of a region can be square or rectangular. However, it can also
               be a circle, oval, triangle, or bloblike.
                   In JPEG 2000, the original image is optionally divided into
               multiple, nonoverlapping subimages called tiles. In case of three
               components (e.g., color composite), all components (e.g., red, green,
               and blue layers) are divided identically.  Alternatively, these com-
               ponents can be linearly combined either reversibly or irreversibly to
               be decorrelated with each other to achieve a high compression ratio.
               The dimension of each tile is always dividable by 2. Arbitrary tile sizes
   126   127   128   129   130   131   132   133   134   135   136