W99681CF Datasheet, PDF (25/78 Pages) Winbond – JPEG USB Dual Mode Camera Chip

W99681CF
7.3 JPEG Compression
The W99681CF supports JPEG baseline sequential process for video data compression. For the
sequential DCT-based mode, 8×8 sample blocks are typically input block by block from left to right, and
block-row by block-row from top to bottom. Each block is transformed by the forward DCT (FDCT) into
a set of 64 values referred to as DCT coefficients. Each of the 64 coefficients is then quantized using
one of 64 corresponding values from a quantization table. After quantization, the DC coefficients and
the 63 AC coefficients are converted into a one-dimensional zig-zag sequence, then are passed to a
Huffman encoder for entropy encoding procedure which compresses the data further.
7.3.1 Level Shift and Forward DCT
Prior to computing the FDCT the input data are level shifted to a signed two′s complement
representation. For 8-bit precision the level shift is achieved by subtracting 128. The following equation
specifies the mathematical definition of the FDCT.
∑ ∑ ^{Svu =}
1
CuCv
7
7
(2x + 1)uπ (2 y + 1)vπ
syx cos
cos
4
x=0 y=0
16
16
where
Cu, Cv = ¹ for u, v = 0
2
Cu, Cv = 1 otherwise
7.3.2 Quantization
After the FDCT is computed for a block, each of the 64 resulting DCT coefficients is quantized by a
uniform quantizer. The uniform quantizer is defined by the following equation. Rounding is to the
nearest integer:
Sqvu
=

round_
Svu 
Qvu 
The quantizer step size for each coefficient S_vu is the value of the corresponding element Q_vu from the
quantization table. The W99681CF supports two programmable quantization tables, luminance
quantization table and chrominance quantization table, which are made by two internal 64×8 SRAMs,
and which should be loaded by the host via the USB bus before start of the JPEG compression.
The quantized DCT coefficient values are signed, two′s complement integers with 11-bit precision for
8-bit input precision.
- 25 -
Publication Release Date: March 2000
Revision A1