Image Compression
08/23/25 Image Compression 1
Reference
[1] Gonzalez and Woods, Digital Image
Processing.
08/23/25 Image Compression 2
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?
08/23/25 Image Compression 3
Enabling Technology
• Compressions is used in
– FAX
– RPV
– Teleconference
– REMOTE DEMO
– etc
08/23/25 Image Compression 4
Review
• What and how to exploit data redundancy
• Model based approach to compression
• Information theory principles
• Types of compression
– Lossless, lossy
08/23/25 Image Compression 5
Information recovery
Data Processing Information
• We want to recover the information, with reduced
data volumes.
• Reduce data redundancy.
• How to measure the data redundancy.
08/23/25 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 = n 1 / n 2
08/23/25 Image Compression 7
Examples
RD = 1 – 1/CR
CR = n 1 / n 2
• 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 ?
08/23/25 Image Compression 8
Types of data redundancy
• Coding
• Interpixel
• Psychovisual
08/23/25 Image Compression 9
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
08/23/25 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.
08/23/25 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
08/23/25 Image Compression 12
Code assignment view
From [1]
08/23/25 Image Compression 13
Interpixel Redundancy
From [1]
08/23/25 Image Compression 14
Run Length Coding
From [1]
CR=1024*343/12166*11
= 2.63
RD = 1-1/2.63 = 0.62
08/23/25 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.
08/23/25 Image Compression 16
From [1]
08/23/25 Image Compression 17
From [1]
08/23/25 Image Compression 18
Fidelity Criteria
• Subjective
• Objective
– Sum of the absolute error
– RMS value of the error
– Signal to Noise Ratio
08/23/25 Image Compression 19
Subjective scale
From [1]
08/23/25 Image Compression 20
Image Compression Model
From [1]
Run length JPEG Huffman
08/23/25 Image Compression 21
