IMAGEDATA COMPRESSION

1

• IMAGE/DATA COMPRESSION
• Need for Data Compression
• When multimedia data objects like documents,
  color or video images are digitized, a large
  amount of digital data are generated. The exact
  amount of data depends an resolution of scanning.
  As resolution increases from 200 dpi to 400 dpi,
  the size of data increases fourfold
  (geometrically).
• Example
• A square inch of 400 dpi image consists of
  160,000 dots ( pixels). If each dot (pixel)
  represent 8 bits of gray level, this information
  becomes 8 x 160 000 bits. An 8-1/2 x 11 inches
  image contains 93.5 sq inch of surface area. An
  uncompressed data object can be of the order of
  several megabytes. These data objects needs to be
  stored, transmitted etc. Calculate!

2

• IMAGE/DATA COMPRESSION
• Need for Data Compression
• The large amount of data present problems in
  storage and transmission.
• Optical media, which has the capability to store
  large volume of data is known to be slower then
  magnetic media.
• Although, network speeds have been increasing,
  and ATM technology is expected to boost speeds to
  well over 100 M-bits/sec, still large data
  objects can take a few seconds to transmit. A
  100Mbits/sec LAN can transmit effectively at
  about half the rate.
• In order to manage large multimedia objects
  efficiently, these objects need to reduce the
  file size.
• Compression algorithms try to eliminate
  redundancies in the pattern data and thus reduce
  the storage required.

3

• IMAGE/DATA COMPRESSION
• Need for Data Compression
• How to reduce data redundancy?
• Example 1 consider a black pixel is followed by
  20 white pixels, there is no need to transmit all
  white pixels.
• LOSS-LESS COMPRESSION Retain all information in
  the multi-media object.
• Compression Standards
• CCITT - Group 2 is a very early compression
  scheme (fax) for 100 dpi black and white images.
• CCITT - Group 3 known as run length encoding.
  This scheme is based on assumption that a typical
  scan line has long run of pixels of the same
  color ( black or white). This scheme is designed
  for black and white images only. The scheme is
  not for gray color images.

4

• Compression Standards
• CCITT Group 3 - 2D compression scheme is also
  known as modified run-length encoding. This
  scheme is more commonly used for software based
  document imaging system.
• While CCITT Group 3- 2D scheme provides fairly
  good compression, it is easier to compress in
  software than CCITT Group 4 standard.The
  compression ratio averages somewhere between
  10-25, between Group 3 and Group 4.
• The compression scheme is based on statistical
  nature of images. For example, the image data
  across the adjacent scan line may normally be
  redundant, if black and white transitions occur
  within plus or minus 3 pixels in the next line as
  well. Depending upon the scan resolution one line
  of text may consist of 20-30 scan lines.

5

• Compression Standards
• Many of these lines have common areas of black
  and white pixels depending upon contour of
  characters. The information that needs to be
  stored is only the information that describes
  changes in the contour of the character, from
  previous line.
• CCITT Group 4 compression standard is a two
  dimensional coding scheme. Only the changes from
  previous line to the next line are transmitted
  using a predefined coding method.
• Group 4 encoding is typically hardware based.
• CCITT - Group 4 does not include shading or color
  information.
• CCITT - Group 5 standard is designed to address
  the need for efficient content-based encoding
  methodology which addresses the color and shade
  information.

6

• Lossy Compression for Photographs and Videos
• The compression technology used for photographs
  are very different from those used for document
  images. Photographs have very high resolution, of
  the order of 1000 pixels per inch. At this
  resolution, uncompressed files are very large.
  The loss of resolution would not have a
  noticeable effect. There are different standards
  - lossy compression schemes.
• Joint Photographic Expert Group ( JPEG) (Part 1
  and 2), formed as a joint ISO CCITT working
  committee, is focused exclusively, known as
  Motion Picture Experts Group (MPEG), is concerned
  with full motion video standards. However, JPEG
  also has standards for gray and color still
  pictures.

7

• Emerging applications such as color fax,
  full-color (24-bit) desk-top publishing, scanners
  and printers need compression standards for data
  reduction that can be implemented at acceptable
  price-performance level.
• JPEG compression standard is designed for still
  as well as moving color and gray-scale
  photographs/ images. The standard has been
  released in two parts.
• Part-I, specifies the modes of operation, the
  interchange formats, and the codec
  specifications for these modes. Part I also
  specifies implementations guidelines.
• Part II of the standard describes the compliance
  tests that determines whether the implementation
  of an encoder or decoder conforms to the
  standards specified in Part I.

8

• While the loss-less compression is always
  desirable, objects with very little information
  redundancy do not produce acceptable results with
  these compression techniques.
• When the compression methods result-in loss of
  some information, the key issue is the effect of
  the loss. Human-eye normally fills-in the missing
  information.
• An important consideration is that significant
  amount of the information should not be lost, and
  human eye (or ear) should not fail to bridge
  information gap in the information.
• Lossy compression is often used for compressing
  audio, gray scale or color images and video
  images in which absolute data accuracy is not
  essential.

9

• A few of the common compression techniques are
• Run-length encoding It is the simplest and
  earliest of data compression scheme. It is
  primarily used to compress black and white
  (binary) images. It is also a basis for other
  types of compression techniques.
• In this scheme, a consecutive repeated string of
  characters is replaced by two bytes. The first
  byte contains a number, representing the of times
  the character is repeated, and the second byte
  contains the character itself.
• 0000 0000 00 1111 111 0000 0000 1111
• 0X0A 0X00 0X07 0X01 0X08 0X00 0X04 0X01

b1 b2 b3 b4 b5 b6 b7 b8
10

• In some cases only one byte can represent the
  pixel value ( 0 or 1 ) and the pixel run length.
  One bit out of 8 bits represent pixel value
  followed by run length.
• The coding of 0 0 1 0 0 1 0 0 represent 0 being
  repeated 36 times.
• This method saves storage space.
• The encoding scheme is carried out on each row
  (one scan line) basis. It does not span across
  multiple lines. Hence it is called a one
  dimensional scheme. The efficiency of this scheme
  is low. However, it is very simple to implement.
  Typical compression efficiencies range from 1/2
  to 1/5. This scheme is included in the TIFF 6.0
  specification..
• If image is changing fast (busy image), the coded
  length

11

• sometimes could be larger than the original
  image (negative compression). Compression
  algorithms should watch out for this effect and
  avoid these.
• CCITT Group 3 1-D Compression Technique
• Assumes that a typical scan-line has long run of
  pixels of the same type (b or w). The scheme is
  for b/w images. It is used for fax transmission
  and software based systems. Run length coding
  discussed above is used.
• Huffman Coding - For Gray Images
• Converts the pixel brightness ( intensity )
  values in the original image to a new variable
  length code based on their frequency of
  brightness values in the image.
• The compression scheme begins by looking at
  brightness histogram of an image.

12

• Huffman Coding
• By ordering the brightness values by their
  frequency of occurrence, a list is created.
• Shortest codes are assigned to most frequent
  values of brightness in the list and the longest
  to the least frequent brightness values
• Consider a 640 x 480 pixel size image with 8
  brightness levels. A histogram is plotted and it
  is found that the intensity values can be
  assigned by 8 brightness values ( in the range
  of 0 - 255 levels).

69,980
67,181
Number of pixels
41,988
34,990
32,891
30,791
27,992
1387
0,0
255
64
128
192
Gray levels
13
Brightness Number of Pixels Huffman code
73 69,989 10 110 67,181
00 146 41,988
110 36 34,990 010 183 32,891
011 219 30,791
1110 255 27,992 11110 0 1,387
11111
0
0
0
307,200
172,138
0
135,062
1
1
0
67,881
102,158
1
1
1
0
60,170
0
1
29,379
1
1
14

• Huffman Coding
• The scheme is very simple to implement in HW/SW
  and is worldwide standard for fax which is
  accepted for document imaging applications.
• The disadvantage is It is 1-D scheme and has
  no error protection scheme.
• Number of bits required for the representation
  will never exceed the number of intensity values.
• If the information content in an image is too
  high (high entropy), the compression may not be
  achieved.
• However, this technique is very good for gray
  images.

15

• CCITT Group 3-2D Compression
• This scheme is known as modified run-length
  coding.
• The scheme is commonly used for software-based
  document imaging systems and facsimile.
• It is easy to decompress and the compression
  ratio of 10 to 20 can be achieved.
• It combines one dimensional coding scheme with
  two dimensional scheme. The 2-D encoding offers
  higher compression because statistically many
  lines differ very little from line above/below.
• It uses k factor, where the image is divided
  into several groups of k lines.

16

• CCITT Group 3-2D Compression
• The first line of every group is encoded using
  CCITT Gr3-1D method. This line becomes the
  reference line for the next line.
• The scheme is based on the statistical nature of
  images the image data across the adjacent scan
  line are redundant.
• If the black and white transition occurs on a
  reference scan line, there are fair chances that
  the same transition will occur within ?3 pixels
  in the next scan line. You code the relative
  transitions.
• In a typical line of text, there may be as many
  as 20-30 scan lines, depending upon the scan
  resolution.

17

• CCITT Group 3-2D Compression
• Many of these scan lines have common areas of
  black and white pixels. The information that
  needs to be stored is only the changes in contour
  of object.
• When this scheme is used, the algorithm embeds
  Group 3, 1D coding( as first line), between every
  k Group 3-2D coding, allowing this to be a
  synchronizing line in the event of a transmission
  error.

0000 0000 00 1111 111 0000 0000 1111 0000 0000 11
1111 111 1000 0011 1111 0X0A 1X07 0X08 1X04 1x02
0x01 1x02
18

• CCITT Group 4-2D Compression
• CCITT Gr 3 2D Standard has been successful.
  However the compression ratio is not too
  impressive (10-20).
• CCITT Gr 4 compression is a 2-D coding scheme, is
  used without a K factor. In this method, the
  reference is first line, which is all white. The
  first group of scan line is coded using the
  imaginary white line.
• The newly coded line becomes reference line for
  the next scan line.
• This provides a high compression ratio. However,
  since the scheme has no reference line(s), a
  single error can result in the rest of the page
  being skewed.
• It is basically the Group 3-2D coding technique
  without the K factor.

19

• Color, Gray scale and Still-Video Image
  Compression
• Emerging applications such as color fax, full
  color desk top publishing, scanners, and printers
  need a compression standard for data reduction
  which can be implemented at acceptable price
  performance levels.
• The existing CCITT Gr 3 and Gr 4 are rally not
  designed for gray/color images and are unable to
  compress sufficiently for viable options.
• The broad JPEG standards are designed to address
  price performance requirements of a variety of
  applications using high resolution graphics.
• The details of JPEG standard are complex and not
  discussed in this course.
• All you have to know on JPEG standard is
  summarized on next few transparencies.

20

• Color, Gray scale and Still-Video Image
  Compression
• JPEG is a compression standard for still color
  images/gray images. It has two parts
• Part-1 Specifies mode of operation, the
  interchange formats, encoder/decoder
  specifications. Implementation guidelines are
  also provided.
• Part-2 Describes compliance tests, which
  determines if the implementation of an
  encoder/decoder conforms to the standard
  specifications to part 1, to ensure
  inter-operability of systems compliant with JPEG
  standards.

21

• Color, Gray scale and Still-Video Image
  Compression
• Requirements for the compression standards also
  include
• Image quality, applicability in any kind of
  continuous tone images, no restriction of image
  dimensions(size), colors, aspect ratios,
  switchover from loss-less to lossy ranges,
  sequential and progressive encoding (image is
  progressively filled with finer details- starting
  with course image).
• Hierarchical Encoding (image is compressed to
  multiple resolution levels, so that lower
  resolution systems can decompress appropriately,
  based on system specifications).Requires multiple
  passes for encoding
• Sequential Encoding entire image is coded in one
  pass- from left to right and top to bottom.

22

• Color, Gray scale and Still-Video Image
  Compression
• JPEG standard has three levels of definitions
• Baseline system
• Extended system
• Special lossless system
• The baseline system reasonably compresses/decompre
  sses color images, maintains a high compression
  ratio, and handle image resolutions from 4-16
  bits/pixel. At this level, the JPEG standard
  ensures that the software implementation, custom
  VLSI implementation and DSP implementation are
  cost effective.

23

• Color, Gray scale and Still-Video Image
  Compression
• The extended system covers the various encoding
  aspects such as variable length encoding,
  progressive encoding and hierarchical mode of
  encoding. This special purpose extensions are
  useful for variety of applications. All these
  methods are extension of baseline sequential
  encoding.
• The special loss-less function ensures that at
  the resolution at which image is compressed, the
  decompression results in no loss of any detail
  that were there in the original source image.
• In other words, there is no loss of details in
  compression/decompression process.

24

• Lossy Image Compression
• Truncation Coding
• During this type of compression technique some of
  the information in the image is lost.
• The technique works by discarding some image data
  using spatial down-sampling and brightness
  resolution-reduction.
• Example Let us consider an image of 640 pixels X
  480 lines, and it has to be printed on a printer
  as 1.5 inches by 1.125 inches using 133 dpi. This
  will translate to 200 pixel x 150 lines image.
• You can reduce the resolution by down sampling -
  regularly dropping pixels and lines
• Similarly brightness levels can also be reduced
  from typically 8 bits (256 levels) to 3 bits (8
  levels) representation.

25
By this brightness compression technique a
reduction of 8/3 2.667 can be achieved.
Original image
4 is to 1 spatial- domain compression- lossy -
one quarter size image
Spatial - truncation coded (8/3) imaged, but
decompressed to full size after truncation
26

• Lossy Image Compression
• You can reconstruct the image coded with reduced
  brightness level with to 3 bits ( from 8
  original bits) by adding additional 5 bits (lsb)
  at receiving side by assigning random brightness
  values ( using random noise pattern generator).
• Another popular form of lossy coding technique
  is
• Transform Coding
• This technique works in frequency domain. First,
  the image is converted into the frequency domain
  using DFT, FFT, etc.
• In the frequency domain, the fundamental
  frequency components represent pixel brightness.
  These components tend to clump in the region
  around low frequency zones.
• In the frequency domain representation, there are
  many high frequency component coefficients with
  very small values

27

• Lossy Image Compression
• This compression technique eliminates high
  frequency coefficients having very small values.
  Normally these components represent high spatial
  frequency contents in the image.
• When the image is inverse transformed back to
  spatial domain, the removal of small valued
  components cause a very little distortion or loss
  of information.
• In the frequency transformed coding, typically, a
  two dimensional image block of 8 x 8 pixel size
  is coded using frequency domain. Only the
  fundamental frequency components are retained and
  stored as the compressed image.
• Decompression operation is merely inverse
  frequency transform of the remaining
  (fundamental) frequency components.

28
Lossy Image Compression
Image compression using Transform Coding . A
compression of 20 1 is achieved, without
significant information loss
29

• Lossy Image Compression
• The quality of the closeness of the decompressed
  image to original image is related by - how
  many coefficients of the frequency domain have
  been discarded?
• Normally the frequency domain representation of
  the image is very efficient form of the image
  representation, and the frequency domain
  compression techniques are very powerful.
• A compression of 20 1 or more can be easily
  achieved on gray level images.
• This technique can be applied to color images
  also.
• Check this operation using Matlab demo command on
  gray level images.

30
Moving Picture Expert Group Standards ( MPEG) H.
261 Uses DCT( discrete cosine transform) and
Huffman coding ( lossy compression technique) It
is used for low quality receptions at 64 k-bits
per seconds transportation links. There are two
more existing standards for moving images MPEG1
and MPEG2 MPEG1 is for relatively lower quality
transmission at 1.5 M bits per seconds
transportation links with resolution of about 320
x 240 pixels . MPEG2 is for higher quality
transmission at 4 - 10 M bits per seconds
transportation links with resolution of about
640 x 480 pixels.