2/22/13

1

Computer Science

CS101 Lecture 13: Image Compression

Vector Graphics Compression Techniques Aaron Stevens (azs@bu.edu) 22 February 2013

Computer Science

What You’ll Learn Today

 Review: how big are image files?  How can we make image files smaller?  What are the main image file formats, and how

are they different from each other?  What are the main techniques for compressing

images?

2/22/13

2

Computer Science

Image File Size: HUGE!

Raster/Bitmap Graphics Storage of data on a pixel-by-pixel basis

  Bitmap (BMP), GIF, JPEG, and PNG, for example

How much data is required to represent a picture?   Typical size might be 1024 by 768 pixels (~ 800,000)   At 3 bytes per pixel, about 2,400,000 bytes for one picture.

  A 10Mpixel picture would be 30,000,000 bytes per picture.

Computer Science

Consider download times…

Example:   A 10Mpixel picture would be 30,000,000 bytes   This is 240,000,000 bits.   Typical home cable modem downloads at 5 Mbits per second

(5,000,000 bits per second).

How long to download this picture?   240,000,000 / 5,000,000 = 48 seconds

Average download speed on 3G networks is closer to 1MB/second. http://www.cnn.com/2011/TECH/mobile/02/18/att.verizon.3g.speed.wired/index.html

2/22/13

3

Computer Science

Describe this picture…

This image was a homework assignment by a CS108 student, Spring 2008.

Vector Graphics describes an image by coordinates, lines, geometric shapes and colors.

Computer Science

Vector Graphics Example

Instructions written in Python by CS108 student (Spring 2010). File size: 29743 bytes.

2/22/13

4

Computer Science

Vector Graphics

Advantages:   Small file sizes (instructions take much less space than

sampling pixels)   Easy to resize mathematically without loss of detail

Disadvantages:

  Image shapes are not lifelike   Color contrast and texturing are rigid, not lifelike   Requires software to interpret the drawing instructions

Computer Science Raster vs. Vector Graphics

When would you use each?

Drawings, Diagrams, Games, etc.

Photography, video, etc.

2/22/13

5

Computer Science Storing an image in fewer bytes What are our choices? What are reasonable tradeoffs?  Raster Graphics vs. Vector Graphics?  Decrease sampling rate (fewer pixels)  Decrease quantization (fewer colors)  Compression algorithms

Computer Science What is the output resolution?   iPhone 3 has a resolution of 640 X 960 pixels.   Evo has 480 X 800 pixels.   Facebook pictures (in web browser) are 560 X 750

pixels.

2/22/13

6

Computer Science

Lower Image Resolution?

Original bitmap image dimensions:   1524 * 2034 = 3,099,816 pixels   3,099,816 pixels * 3 bytes = 9,299,448 bytes

How about 300 * 400 pixels?   300 * 400 pixels = 120,000 pixels   120,000 pixels * 3 bytes = 360,000 bytes

A compression ratio of 0.038. What’s the catch?

Computer Science

Fewer Colors: GIF Images

Graphics Interchange Format (GIF) Each image is made up of any 256 (or 16) RGB colors, but only those colors.   256 colors: 8 bits per pixel   16 colors: 4 bits per pixel

2/22/13

7

Computer Science

16 Color GIF 120,000 bytes

256 Color GIF 200,000 bytes

Original BMP 360,000 bytes

Fewer Colors: GIF Images

Computer Science

Compress them! How do you fit more cars on this lot?

Data compression Reduction in the amount of space needed to store a piece of data.

2/22/13

8

Computer Science

Data Compression (saving space)

Data compression Reduction in the amount of space needed to store a piece of data. Data compression techniques can be:

lossless, which means the data can be retrieved without any loss of the original information

lossy, which means some information may be lost in the process of compaction

Computer Science

JPEG Images

The Joint Photographic Experts Group created the standard codec in 1992.  Default format on most digital cameras  Compression reduces file size up to 90% with

little loss in visible quality  Great for life-like images the web

Is JPEG compression lossy or lossless?

2/22/13

9

Recall: RGB Colorspace

=

RGB has a lot of redundant information.

JPEG uses the YCbCr Colorspace

Y (luminance)

Cb (chroma blue)

Cr (chroma red)

2/22/13

10

Computer Science YCbCr Colorspace

YCbCr separates luminance (brightness) from chroma (color differences).  Our eyes are more sensitive to changes in luma

than to changes in chroma.

JPEG thus stores a high resolution Y, and lower-resolution CbCr.   Smoothing colors among adjacent pixels.

Computer Science

JPEG Color Smoothing

This image is progressively more less smoothed from left to right (lowest to highest quality).

2/22/13

11

Computer Science

JPEG Compression

 Convert color space from RGB to YCbCr   Smooth adjacent colors in Cb/Cr   Apply Huffman compression after smoothing

Which part of this is lossy compression?

Computer Science

PNG Image Format

Portable Network Graphics (PNG) 24-bit bitmapped color. Uses lossless compression.

2-stage compression process: 1.  Filter image by adjacent pixels,

record the differences 2.  Use the DEFLATE algorithm

(Huffman encoding scheme)

2/22/13

12

Computer Science Same Picture, Many Formats Original image was

2048 * 1536 pixels All others are

300 * 400 pixels

Computer Science

2/22/13

13

Computer Science

What You Learned Today

  Vector graphics vs. raster graphics   Factors in image file size

  Resolution, color depth

 Compression strategies   Reduce colors   Reduce resolution   Lossless compression algorithms   Lossy compression algorithms

 GIF, JPEG, and PNG

Computer Science

Announcements and To Do

 HW06 due Wednesday 2/27  Readings:

  Wong ch 3, pp 66-86 (today)   Wong ch 4, pp 102-117 (next week)

2/22/13

14

Computer Science Image Manipulation Software

There are many different image manipulation software packages, e.g.:   Paint – Built in to windows   Photoshop – Commercial software   Irfanview – Free software   Paintshop Pro – Shareware/Commercial  GIMP – Free software

Computer Science

Example: Text Compression

Problem: Unicode assigns 16 bits to each character in a document; uses a heck of a lot of space. We need ways to store and transmit text efficiently. Why? Common lossless compression techniques:

keyword encoding run-length encoding Huffman encoding

2/22/13

15

Computer Science

Keyword Encoding

Replace frequently used words with a single character

Computer Science

Keyword Encoding Example

Given the following paragraph, We hold these truths to be self-evident, that all men are created equal, that they are endowed by their Creator with certain unalienable Rights, that among these are Life, Liberty and the pursuit of Happiness. That to secure these rights, Governments are instituted among Men, deriving their just powers from the consent of the governed, That whenever any Form of Government becomes destructive of these ends, it is the Right of the People to alter or to abolish it, and to institute new Government, laying its foundation on such principles and organizing its powers in such form, as to them shall seem most likely to effect their Safety and Happiness.

2/22/13

16

Computer Science

Keyword Encoding Example

The encoded paragraph is We hold # truths to be self-evident, $ all men are created equal, $ ~y are endowed by ~ir Creator with certain unalienable Rights, $ among # are Life, Liberty + ~ pursuit of Happiness. — $ to secure # rights, Governments are instituted among Men, deriving ~ir just powers from ~ consent of ~ governed, — $ whenever any Form of Government becomes destructive of # ends, it is ~ Right of ~ People to alter or to abolish it, + to institute new Government, laying its foundation on such principles + organizing its powers in such form, ^ to ~m shall seem most likely to effect ~ir Safety + Happiness.

Computer Science

Keyword Encoding

Compression ratio The size of the compressed data divided by the size of the original data (0 < c.r. <= 1) What did we save?

Original paragraph: 656 characters Encoded paragraph: 596 characters Characters saved: 60 characters Compression ratio: 596/656 = 0.9085

Could we use this substitution chart for all text? How could we apply keyword encoding to images?

2/22/13

17

Computer Science

Run-Length Encoding

Consider a single character which is repeated over and over again in a long sequence. Replace a repeated sequence with

  a flag character   repeated character   number of repetitions

Example: *n8   * is the flag character   n is the repeated character   8 is the number of times n is repeated

Computer Science Run-Length Encoding Example

Original text bbbbbbbbjjjkllqqqqqq+++++

Encoded text *b8jjjkll*q6*+5 (Why isn't l encoded? J?)

The compression ratio is 15/25 or .6

Encoded text *x4*p4l*k7

Original text xxxxpppplkkkkkkk

This type of repetition isn’t very helpful for English text. Can you think of a situation where it might be helpful?

2/22/13

18

Computer Science Run-Length Encoding Example

How would you describe the colors in this image?

Computer Science Which letters occur most?

2/22/13

19

Computer Science

Huffman Encoding

Why should the character “X" and "z" take up the same number of bits as "e" or " "? Huffman codes use variable-length bit strings to represent each character.

More frequently used letters have fewest bits.

Computer Science

Huffman Encoding Example

ballboard would be 1010001001001010110001111011!

!

Encoded is 28 bits vs 144 bits with Unicode; The compression ratio is 28/144 or 0.39 Try to encode roadbed

2/22/13

20

Computer Science

Huffman Encoding

Prefix Property No character's bit string is the prefix of any other character's bit string. To decode

look for match left to right, bit by bit record letter when a match is found begin next character where you left off

Computer Science

Huffman Encoding Example

Decode!!1011111001010!

Try it!

2/22/13

21

Computer Science

Huffman Encoding

The technique for creating codes guarantees the prefix property of the codes. There is no single “Huffman code” -- each depends on the application. Two types of Huffman codes:

  general, based on use of letters in English, Spanish, ….   specialized, based on text itself or specific types of text