Image Compression

7/14/2012

Image Compression

Reference
[1] Gonzalez and Woods, Digital Image Processing.

7/14/2012

Image Compression

Objective
Reduce the number of bytes required to represent a digital image
Redundant data reduction Remove patterns Uncorrelated data confirms redundant data elimination
Auto correlation?

7/14/2012

Image Compression

Enabling Technology
Compressions is used in
FAX RPV Teleconference REMOTE DEMO etc

7/14/2012

Image Compression

Review
What and how to exploit data redundancy Model based approach to compression Information theory principles Types of compression
Lossless, lossy

7/14/2012

Image Compression

Information recovery
Data Processing Information

We want to recover the information, with reduced data volumes. Reduce data redundancy. How to measure the data redundancy.
7/14/2012 Image Compression 6

Relative Data Redundancy

Assume that we have two data sets D1 and D2.
Both on processing yield the same information. Let n1 and n2 be the info carrying units of the respective data sets. Relative data redundancy is defined on comparing the relative dataset sizes

RD = 1 1/CR where CR is the compression ratio CR = n1 / n2

7/14/2012 Image Compression 7

Examples
RD = 1 1/CR CR = n1 / n2 D1 is the original and D2 is compressed. When CR = 1, i.e. n1 = n2 then RD=0; no data redundancy relative to D1 . When CR = 10, i.e. n1 = 10 n2 then RD=0.9; implies that 90% of the data in D1 is redundant. What does it mean if n1 << n2 ?
Image Compression 8

7/14/2012

Types of data redundancy

Coding Interpixel Psychovisual

7/14/2012

Image Compression

Coding Redundancy
How to assign codes to alphabet In digital image processing
Code = gray level value or color value Alphabet is used conceptually

General approach
Find the more frequently used alphabet Use fewer bits to represent the more frequently used alphabet, and use more bits for the less frequently used alphabet
7/14/2012 Image Compression 10

Coding Redundancy 2
Focus on gray value images Histogram shows the frequency of occurrence of a particular gray level Normalize the histogram and convert to a pdf representation let rk be the random variable
pr(rk) = nk/n ; k = 0, 1,2 ., L-1, where L is the number of gray level values l(rk) = number of bits to represent rk Lavg = k=0 to L-1 l(rk) pr(rk) = average number of bits to encode one pixel. For M x N image, bits required is MN Lavg For an image using an 8 bit code, l(rk) = 8, Lavg = 8. Fixed length codes.
7/14/2012 Image Compression 11

Fixed vs Variable Length Codes

From [1]

Lavg = 2.7 CR= 3/2.7 = 1.11 RD = 1 1/1.11 = 0.099

7/14/2012 Image Compression 12

Code assignment view

From [1]

7/14/2012

Image Compression

13

Interpixel Redundancy

From [1]

7/14/2012

Image Compression

14

Run Length Coding

From [1]

CR=1024*343/12166*11 = 2.63 RD = 1-1/2.63 = 0.62

7/14/2012

Image Compression

15

Psychovisual Redundancy
Some visual characteristics are less important than others. In general observers seeks out certain characteristics edges, textures, etc and the mentally combine them to recognize the scene.

7/14/2012

Image Compression

16

From [1]

7/14/2012

Image Compression

17

From [1]

7/14/2012

Image Compression

18

Fidelity Criteria
Subjective Objective
Sum of the absolute error RMS value of the error Signal to Noise Ratio

7/14/2012

Image Compression

19

Subjective scale

From [1]

7/14/2012

Image Compression

20

Image Compression Model

From [1]
Run length JPEG Huffman

7/14/2012

Image Compression

21