Many of the JPEG2000 are related to the pre-processing and segmentation of the image. The general color image has three components such as RGB, or YUV, etc. Before the primary image segmentation is performed, the original image signal is now dedicated (so-called DC layer) and the component transformation is performed (mainly to convert the RGB to YCBCR). Only one component or black and white image of the color image is discussed here.
The first important concept is to refer to the grid. This is equivalent to the basic coordinate system of the JPEG2000 image. The image area is equivalent to an offset under the coordinate system (generally referred to (xosiz, yosiz)), and is at the end of an offset (generally recorded (xsiz-1, ysiz-1)) Rectangular area. For different components, different secondary sampling may be used, for example, the brightness component sampling of YCBCR is generally sampled high. The sampling intervals in the X and Y directions are: XRSIZ (C) and YRSIZ (C). The image area divides the rectangular mesh that is filled with blocks into a plurality of blocks. Fill the block split grid starting point (Xtosiz, Ytos), the size of the grid is (Xtsiz, Ytsiz). Naturally, you must meet: 0 <= xtosiz <= xosiz, 0 <= Ytosiz <= Yosiz, Xtosiz XSiz> xosiz, Ytosiz Ytsiz> Yosiz, so that the grid of the uppermost corner contains the first image segmentation Piece. The index of the block is filled in the raster order. Fill the block segmentation Similar to the encoding segmentation in the JPEG, and the latter is fixed.
For the components of different secondary sampling frequencies, although the filling block segmentation is the same in the reference mesh, the number and number of sample points and numbers in the same padding point are different. Specific conversion is not complex, omitted here.
Although the core work of JPEG2000 is done in each filled block, but the coordinates of the sampling point fill the faster (not referred to the absolute coordinates on the grid, but the sampling index in the X and Y directions) It is sensitive, this coordinates should be used as parameters for wavelet transform. The subband generally records NBXBY, where N is 0 or positive integer, indicating the decomposition level of the sub-band, BX and BY Take L or H, indicate low pass and high pass, respectively. Each NLL subband has a resolution level, represented by R, R = NL-N, NL is the highest decomposition level. Each subband has a memory interval to save information on the subband. Thereby, decomposing the subband filling block is to fill in the corresponding position on the storage section of each subband, and the entire image is performed as the filled block operation is performed until the entire image is performed. The synthesis is an opposite process.
The information of each subband is subjected to sub-zone segmentation, and the sub-zone length is limited to 2 power.
(Endlessly)
