Data Compression

Understanding Data Communications and Networks William A. Shay (3rd Edition)

Introduction
Fax machine: 40000 DPSI => 4 million dots per page 56 KBPS modem, time to transmit = ? Video: 30 pictures per second Each picture = 200,000 dots or pixels 8-bits to represent each primary color Bits required for one picture = ? Two hour movie requires = ?

Introduction
Compression is a way to reduce the number of bits in a frame but retaining its meaning. Decreases space, time to transmit, and cost Technique is to identify redundancy and to eliminate it If a file contains only capital letters, we may encode all the 26 alphabets using 5-bit numbers instead of 8-bit ASCII code

Introduction
If the file had n-characters, then the savings = (8n-5n)/8n => 37.5%

Frequency Dependent Codes

Not all the characters appear with same frequency, some are more prevalent than the others Frequently appearing characters could be assigned shorter codes than the others => results in reduced number of bits Such codes are examples of frequency dependent code

Frequency Dependent Codes

Huffman code: (illustrated with a manageable example) Letter Frequency (%) A 25 B 15 C 10 D 20 E 30

Frequency Dependent Codes

Huffman code: Code formation - Assign weights to each character - Merge two lightest weights into one root node with sum of weights (if multiple? Not unique code) - Repeat until one tree is left - Traverse the tree from root to the leaf (for each node, assign 0 to the left, 1 to the right)

Frequency Dependent Codes

Huffman code: Code Interpretation - No prefix property: code for any character never appears as the prefix of another code (Verify) - Receiver continues to receive bits until it finds a code and forms the character - 01110001110110110111 (extract the string)

Frequency Dependent Codes

Arithmetic compression: is based on Interpreting a character-string as a single real number Letter Frequency (%) Subinterval [p, q] A 25 [0, 0.25] B 15 [0.25, 0.40] C 10 [0.40, 0.50] D 20 ? E 30 ?

Frequency Dependent Codes

Arithmetic compression: Coding CABAC Generate subintervals of decreasing length, subintervals depend uniquely on the strings characters and their frequencies. Interval [x, y] has width w = y x, the new interval based on [p, q] is x = x + w.p, y = x + w.q Step 1: C 0..0.4.0.5..1 based on p = 0.4, q = 0.5

Frequency Dependent Codes

Step 2: A 0.40.425.....0.5 based on p = 0.0, q = 0.25 Step 3: B 0.40.406250.41..0.425 based on p = 0.25, q = 0.4 Step 4: A Step 5: C 0.406625 0.4067187 Final representation (midpoint)?

Frequency Dependent Codes

Arithmetic compression: Extracting CABAC
N 0.4067 0.067 0.268 ? ? Interval[p, q] 0.4 0.5 0 0.25 0.25 0.4 Width 0.1 0.25 0.15 Character C A B N-p 0.0067 0.067 0.018 (N-p)/width 0.067 0.268 0.12

When to stop? A terminal character is added to the original character set and encoded. During decompression, once it is encountered the process stops.

Run Length Encoding

Huffman code requires: - frequency values - bits are grouped into characters or units Many items do not fall into such category - machine code files - facsimile Data (bits corresponding to light or dark area of a page) - video signals

Run Length Encoding

For such files, RLE is used. Instead of sending long runs of 0s or 1s, it sends only how many are in the run. 70%-80% space is white on a typed character space, so RLE is useful.
A

Run Length Encoding

Runs of the same bit In facsimile Data, there are many 0s (white spots) > transmit the run-length as fixed size binary integer Receiver generates proper number of bits in the run and inserts the other bit in between 14 zeros, 1, 9 zeros, 11, 20 zeros, 1, 30 zeros, 11, 11 zeros (number of zeros encoded in 4bits)

Run Length Encoding

Runs of the same bit Code: 1110 1001 0000 1111 0101 1111 1111 0000 0000 1011 (next value after 1111 is added to the run) SAVINGS IN BITS: ? If the stream started with 1 instead? Best when there are many long runs of zeros, with increased frequency of 1s, becomes less efficient.

Run Length Encoding

Runs with different characters Send the actual character with the run-length HHHHHHHUFFFFFFFFFYYYYYYYYYYYDGGGGG code = 7, H, 1, U, 9, F, 11, Y, 1, D, 5, G

SAVINGS IN BITS (considering ASCII): ?

Run Length Encoding

Facsimile Compression ITU standard (A4 document, 210 by 297 mm) 1728 pixels per line If 1 bit for each pixel, then over 3 million bits for each page A typical page contains many consecutive white or black pixels -> RLE

Run Length Encoding

Run lengths may vary from 0 to 1728 -> many Possibilities and inefficiency with a fixed size code Some runs occur more frequently than others, e.g. most typed pages contain 80% white pixels, spacing between letters is fairly consistent => probabilities of certain runs are predictable => Frequency dependent code on run lengths

Run Length Encoding

Some Facsimile compression codes (Terminating, less than 64) Pixels in the run Code: White Code: Black 0 00110101 0000110111 1 000111 010 2 0111 11 3 1000 10 10 00111 0000100 20 0001000 00001101000

Run Length Encoding

Some Facsimile compression codes (Make up, greater than or equal to 64) Pixels in the run Code: White Code: Black 64 11011 0000001111 128 10010 000011001000 256 512 # 129 white: Savings:
No-prefix property, better compression for long-runs

Relative Encoding
Relative Encoding: Some applications may not benefit from the above: video image -> little repetitive within, but much repetition from one image to the next Differential encoding is based on coding only the difference from one to the next

Relative Encoding
Relative Encoding: 1234 1334 2537 2537 3648 3647 4759 3759 1st Frame 2nd Frame Resulting difference can be RLE.

0100 0000 0 0 0 -1 -1 0 0 0 Difference

Image Representation
BW pixels each represented by 8-bit level Color composed of R, G, B primaries, each is represented by 8-bit level -> Each color pixel can be represented by one of 28 .28.28 = 224 colors VGA screen: 640 * 480 pixels -> 640 * 480 * 24 = 7, 372, 800 bits

Image Compression
JPEG compression both for grayscale and color images Previous compression methods were lossless it was possible to recover all the information from the compressed code JPEG is lossy: image recovered may not be the same as the original

JPEG Compression
It consists of three phases: Discrete Cosine Transform (DCT), Quantization, Encoding. DCT: Image is divided into blocks of 8*8 pixels For grey-scale images, pixel is represented by 8-bits For color images, pixel is represented by 24bits or three 8-bit groups

JPEG Compression
DCT takes an 8*8 matrix and produces another 8*8 matrix. T[i][j] = 0.25 C(i) C(j) P[x][y] Cos (2x+1)i/16 * Cos (2y+1)j/16 i = 0, 1, 7, j = 0, 1, 7 C(i) = 1/2, i =0 = 1 otherwise T contains values called Spatial frequencies

JPEG Compression
Spatial frequencies directly relate to how much the pixel values change as a function of their positions in the block T[0][0] is called the DC coefficient, related to average values in the array/matrix, Cos 0 = 1 Other values of T are called AC coefficients, cosine functions of higher frequencies

JPEG Compression
Case 1: all Ps are same => image of single color with no variation at all, AC coefficients are all zeros. Case 2: little variation in Ps => many, not all, AC coefficients are zeros. Case 3: large variation in Ps => a few AC coefficients are zeros.

JPEG Compression
P-matrix T-matrix 20 30 40 50 60 720 -182 0 -19 0 30 40 50 60 70 -182 0 0 0 0 40 50 60 70 80 0 0 0 0 0 50 60 70 80 90 -19 0 0 0 0 60 70 80 90 100 0 0 0 0 0 Uniform color change and little fine detail, easier to compress after DCT

JPEG Compression
P-matrix T-matrix 100 150 50 100 100 835 15 -17 59 5 200 10 110 20 200 46 -60 -36 11 14 10 200 130 30 200 -32 -9 130 105 -37 100 10 90 190 120 59 -3 27 -12 30 10 200 200 120 90 50 -71 -24 -56 -40 . . Large color change, difficult to compress after DCT

JPEG Compression
To restore, inverse DCT (IDCT) is performed: P[x][y] = 0.25 C(i) C(j) T[i][j] Cos (2x+1)i/16 * Cos (2y+1)j/16 x = 0, 1, 7, y = 0, 1, 7 # Can write a C-program to apply DCT on a Parray (8*8) to obtain T-array and also IDCT on T-array to recover P-array.

JPEG Compression
Quantization: Provides an way of ignoring small differences in an image that may not be perceptible. Another array Q is obtained by dividing each element of T by some number and roundingoff to nearest integer => loss

JPEG Compression
T-matrix Q-matrix 152 0 -48 0 -8 15 0 -5 0 -1 0 0 0 0 0 0 0 0 0 0 -48 0 38 0 -3 -5 0 4 0 0 0 0 0 0 0 0 0 0 0 0 -8 0 -3 0 13 -1 0 0 0 1 Divide by 10 and round-off => creates fewer distinct numbers and more consistent pattern

JPEG Compression
Can we recover by multiplying the elements with 10? If the loss is minimal, the vision system may not notice. Dividing T-elements by the same number is not practical, may result in too much loss. Retain the effects of lower spatial frequencies as much as possible less subtle features noticed if changed

JPEG Compression
Upper left elements to be divided with smaller values Lower right elements higher spatial frequencies, finer details - to be divided with larger values To define a quantization array U, then Q[i][j] = Round (T[i][j] U[i][j]), i = 0, 1, 7, j = 0, 1, 7

JPEG Compression
U = 1 3 5 7 9 Q= 152 0 -10 0 -1 3 5 7 9 11 0 0 0 0 0 5 7 9 11 13 -10 0 4 0 0 7 9 11 13 15 0 0 0 0 0 9 11 13 15 17 -1 0 0 0 0 Q can be well-compressed due to redundant elements

JPEG Compression
Encoding - linearize two dimensional data and compress it 152 0 -10 0 -1 0 0 0 0 0 -10 0 4 0 0 0 0 0 0 0 -1 0 0 0 0 - Row-wise (shorter runs) - Zigzag (longer runs, higher spatial frequencies are gathered together)

JPEG Compression
Many such 8*8 arrays, adjacent blocks with little difference => more potential for compression JPEG may provide 95% compression (depends on image and quantization array) GIF (Graphic Image Format) reduces color to 256. Best suited for a few colors and sharp boundaries (charts, lines). Not good for variations and shading full color photo

Multimedia Compression
JPEG compresses still pictures, motion pictures are compressed by MPEG. Still picture contains 7, 372,800 bits. Compression ratio 20:1 reduces the bits to 368,640 With 30 images per second, bits to be handled 11,059,200. => Huge data transmission if channel is shared

Multimedia Compression
Based on JPEG compression and relative encoding Usually little difference from one to the next frame => suitable for compression A completely new scene can not be compressed this way. Also, not suited for moving objects unleashing other objects. Three frames: I, P, B.

Multimedia Compression
I (intra-picture) - frame: Just a JPEG encoded image. P (predicted) frame: Encoded by computing the differences between a current and a previous frame. B (bidirectional) - frame: Similar to P-frame except that it is interpolated between previous and future frame.

Multimedia Compression
I-frames must appear periodically in any frame sequence, otherwise: - any error introduced in any frame is propagated in all subsequent frames if relative differences are sent - in broadcast applications, if one tunes-in late, then he has nothing to compare to

Multimedia Compression
Typical sequence: order in which to transmit: I P I P is sandwiched between groups of B P is the difference from the previous I-frame B is interpolated from the nearest I and P order in which to display I B B P B B I

Multimedia Compression
Coding P: Motion compensated Prediction for P-frames Divide the image into macro-blocks of 256 pixels (16*16) Chrominance arrays are reduced to groups of 4 pixels (becomes 8*8 matrix) representing average values => loss, not perceptible

Multimedia Compression
An algorithm examines each macro-block of Pframe and locates a best matching macroblock in the prior I-frame Not necessarily in the same relative position (in the vicinity) Once located, the algorithm calculates the difference and motion vector (displacement) The above information is encoded and transmitted

Multimedia Compression
At the decoding end, the motion-vector is used to determine the position and the difference is used to reconstruct the macroblock

Multimedia Compression
MP3 (MPEG Level 3) compression protocol for audio Audible range: 20 20 KHz PCM uses 16 bit samples @ 44.1 KHz 1-sec of PCM audio needs: 16*44.1*1000 bits For two channel stereo: Twice For 1-minute audio: 60 times -> 88 Mbits

Multimedia Compression
MP3 is based on a psychoacoustic model what we can hear and what we can distinguish Auditory masking capture an audio signal, determine what we can not hear, remove those components from the stream, digitize what is left Sub-band coding decompose the original signal into non-overlapping frequency ranges, create bit-stream for each band

Multimedia Compression
Sub-bands are encoded differently: Sub-bands with loud signals need good resolution, are coded with more bits Near-by sub-bands with weak signals are effectively masked by louder signals, need less resolution, are coded with fewer bits, that results in compression