Main steps in JPEG image
step1. Transform RGB to YUV and subsampe color
step2. DCT on images blocks (DC,AC)
step3. Quantization(low frequency-small step, high frequency-big step)(The quantization step of luminance（亮度） signal is longer than that of chroma signal.）
step4. Zigzag ordering and run-length encoding and DPCM
Zigzag order and Run-length coding on AC coefficients
start:lowest coefficiect
end:highest coefficient
To increase the number of the zeros.
{zeros-to-skip,next non-zero value}(end tag:(0,0))-for AC coefficient
for example{41,0,-2,0,0,0,0,0,5,0,0,0,0,0...} can be represented as {(0,41),(1,-2),(5,5)(0,0)}
DPCM:difference code(差值编码)-for DC coefficient
for example: 150,155,149,152 can be coded as 150,5,-6,3...
Huffman coding ALGORITHM: FREQUENCY-occurrence probability-for DC coefficient
Huffman coding method: the combination of the value from the root to each node.
for example: value:probabilty
1:3/16,2:5/16,3:2/16,4:1/16,5:3/16,6:2/16
start from the value of small probability, finally compare the length of the code.
compression ratio = code_length(uncompression)/code_length(compression)
average code length = add_all(probability*each_code_length.
multiple coding results, but unique decoding and unique average code length.
Entropy Encoding: (size, value):(for nagetive, we use its anti-code)
for example, codes : 150 5 -6-> (8 10010110),(3,101),(3,001)
size is Huffman coded. While value is not Huffman coded.
Extension: TLV->type,length, value
AC Huffman coding: (runlength,value)->runlength=(runlength(4-bit),size(4-bit)); value=(amplitude(8-bit))
step5. Entopy encoding
MODE OF JPEG
Sequential mode: a single order from right to left, from top to bottom
Progressive mode: delivers low quality versions of the image quickly, followed by higher quality passses.
step1: DC + low frequency AC coefficients
step2: higher frequency AC coefficients
step3: highest frequency AC coefficients
