www hndit com Image Compression 1 www hndit

www. hndit. com Image Compression 1

www. hndit. com What is an Image File Format? � Image file formats are standard way of organizing and storing of image files. � Image files are composed of either pixels or vector (geometric) data. � The pixels that compose an image are ordered as a grid (columns and rows) � Each pixel consists of numbers representing levels of brightness and color. 2

www. hndit. com Image File Size � This is expressed as number of bytes � This value depends on two factors ◦ Resolution (Number of pixels on the screen) ◦ Bit depth (Number of bits allocated for a pixel) � Large number of rows and columns increase the resolution and it leads to larger the file size. 3

www. hndit. com Image File Compression � Compression is a term used to describe ways of cutting the size of the file. 4

www. hndit. com Why Need Image Compression? � Image files typically are larger than text files. � Web pages often contain many images that are transmitted across slow connections � A larger file type means more disk usage and slower downloads Therefore it is helpful to have a way to represent images in a compact format. 5

www. hndit. com Types of Image File Compression � There are two types of Compression algorithms. Lossless Compression Lossy Compression 6

www. hndit. com Lossless Compression � Algorithms reduce file size without losing image quality � They are not compressed into as small a file as a lossy compression file. � When image quality is valued than file size, lossless algorithms are typically chosen. �because it lets you recreate the original file exactly. � All lossless compression is based on the idea of breaking a file into a "smaller" form for transmission or storage and then putting it back together on the other end so it can be used again. 7

www. hndit. com Lossy compression � This algorithms consider the limitations of the human eye and discard invisible information. � Most lossy compression algorithms allow for variable quality levels (compression) � As these levels are increased File size is reduced. � At the highest compression levels, image weakening becomes noticeable � Lossy compression works very differently. 8

www. hndit. com Lossy compression… � These programs simply eliminate "unnecessary" bits of information, so that it is smaller. � This type of compression is used a lot for reducing the file size of bitmap pictures, which tend to be fairly large. 9

www. hndit. com Lossy compression… Example: � If there is an image where whole sky is blue, most of the individual pixels are a little bit different. � To make this picture smaller without compromising the resolution, you have to change the color value for certain pixels. � If the picture had a lot of blue sky, the program would pick one color of blue that could be used for every pixel. 10

www. hndit. com Lossy compression… � Then, the program rewrites the file so that the value for every sky pixel refers back to this information. � If the compression scheme works well, you won't notice the change, but the file size will be significantly reduced. � with lossy compression, you can't get the original file back after it has been compressed. 11

www. hndit. com Compression Algorithms Algorithm Basic Concept Comp Ratio File Format Loss-Less Lossy RLE (Run-Length Encoding) Compress repetitive data ~1. 2 BMP LZW (Lempel-Ziv-Welch) Build treed dictionary ~2. 0 TIFF, GIF DCT (Discrete Cosine Transformation Transform to series of Cosine functions ~100 JPEG, MPEG 1/2 Colour Space Compression Cut non-sensitive color information ~2 JPEG, (TV) Wavelet Transform to series of Wavelet functions ~100 JPEG 2000, MPEG 4 12

www. hndit. com Some General Concepts • Bit Rate and Compression Ratio – Bit rate: bits/pixel, sometimes written as bpp – Compression ratio (CR): CR = number of bits to represent the original image number of bits in compressed bit stream 13

Run Length Encoding (RLE) www. hndit. com � Many files, particularly image files, contains sequences of identical symbols. ◦ Eg. In an image, a section of many adjacent pixels may all be the same color. ◦ Be encoded with the same bit pattern. � RLE replaces sequence of identical bit patterns with ◦ one instance of the pattern, and ◦ a number specifying how many times the pattern is to be repeated. � Uses with BMP 14

www. hndit. com • Run Length Encoding – The length of consecutively identical symbols • Run length encoding Example • When Does it Work? – Images containing many runs of 1’s and 0’s • When Does it Not Work? 15

www. hndit. com Run Length Encoding • Decoding Example A binary image is encoded using run length code row by row, with “ 0” represents white, and “ 1” represents black. The code is given by Row Row Row Row 1: 2: 3: 4: 5: 6: 7: 8: 9: 10: 11: 12: 13: 14: 15: 16: “ 0”, 16 “ 0”, 7, 2, 7 “ 0”, 4, 8, 4 “ 0”, 3, 2, 6, 3, 2 “ 0”, 2, 2, 8, 2, 2 “ 0”, 2, 1, 10, 1, 2 “ 1”, 3, 10, 3 “ 0”, 2, 1, 10, 1, 2 “ 0”, 2, 2, 8, 2, 2 “ 0”, 3, 2, 6, 3, 2 “ 0”, 4, 8, 4 “ 0”, 7, 2, 7 “ 0”, 16 decode Decode the image 16

www. hndit. com Run Length Encoding • Decoding Example A binary image is encoded using run length code row by row, with “ 0” represents white, and “ 1” represents black. The code is given by Row Row Row Row 1: 2: 3: 4: 5: 6: 7: 8: 9: 10: 11: 12: 13: 14: 15: 16: “ 0”, 16 “ 0”, 7, 2, 7 “ 0”, 4, 8, 4 “ 0”, 3, 2, 6, 3, 2 “ 0”, 2, 2, 8, 2, 2 “ 0”, 2, 1, 10, 1, 2 “ 1”, 3, 10, 3 “ 0”, 2, 1, 10, 1, 2 “ 0”, 2, 2, 8, 2, 2 “ 0”, 3, 2, 6, 3, 2 “ 0”, 4, 8, 4 “ 0”, 7, 2, 7 “ 0”, 16 decode 17

www. hndit. com Chain Coding Assume the image contains only single-pixel-wide contours, like this, not this contour image After the initial point position, code direction only region image n 2 3 1 0 4 Code Stream: 7 (3, 2), 1, 0, 1, 1, 3, 3, 3, 4, 4, 5, 4 initial point position chain code 5 m 6 18

www. hndit. com Chain Coding • Decoding Example The chain code for a 8 x 8 binary image is given by: 2 3 (1, 3), 7, 7, 0, 1, 1, 3, 3, 3, 1, 1, 0, 7, 7 decode 1 4 0 5 6 7 Decode the image 19

www. hndit. com Chain Coding • Decoding Example The chain code for a 8 x 8 binary image is given by: 2 3 (1, 3), 7, 7, 0, 1, 1, 3, 3, 3, 1, 1, 0, 7, 7 decode 1 4 0 5 6 7 20

Variable Word Length Coding www. hndit. com Assign short words to gray levels that occur frequently Assign long words to gray levels that occur infrequently Example: A 4 x 4 4 bits/pixel original image is given by Default Code Book 0: 1: 2: 3: 4: 5: 6: 7: 8: 9: 10: 11: 12: 13: 14: 15: 0000 0001 0010 0011 0100 0101 0110 0111 1000 1001 1010 1011 1100 1101 1110 1111 Bit rate = 4 bits/pixel Total # of bits used to represent the image: encode 4 x 16 = 64 bits 21

www. hndit. com Variable Word Length Coding: Example • Encode the original image with a CODE BOOK given left Huffman Code Book 0: 1: 2: 3: 4: 5: 6: 7: 8: 9: 10: 11: 12: 13: 14: 15: 00000001 0000010 0000011 0000100 01 0000101 10 00100 11 0000110 0000111 001010 0011 001011 Total # of bits used to represent the image: encode 4+2+2+2+2+2+5+2+2+4 = 39 bits Bit rate = 39/16 = 2. 4375 bits/pixel CR = 64/39 = 1. 6410 22

www. hndit. com Lempel-Ziv-Welch (LZW) � Dictionary based coding algorithm � Another Loss-Less compression algorithm. � It was not designed specifically for graphics � Data Dictionary is used to represent linear sequences of data in a uncompressed input stream. Then uses an algorithm similar to RLE. � It does not work well with black and white or true colour images. � Uses with GIF 23

www. hndit. com Color Space Compression � Uses human eye characteristics ◦ Less sensitive to color than lightness ◦ Less sensitive to Red than Green � YUV color space ◦ Originally developed for color TV signal ◦ Convert color to Luminance(Y) and � Chrominance (U, V) values Y=0. 299 R + 0. 587 G + 0. 114 B … Most sensitive color for human eyes U=(B-Y) V=(R-Y) 24

www. hndit. com Color Space Compression cont… � Reduce color information § Y: U: V = 4: 2: 2 (TV) § Y: U: V = 4: 1: 1 (JPEG) § Y: U: V = 4: 1: 0 (JPEG) JPEG (1: 50) Y U V 25

www. hndit. com Mathematical Transformation � Convert images to mathematical functions ◦ Discrete Cosine Transformation (DCT) �Use series of cosine functions to approximate image. �Use with JPEG, MPEG 1/2 ◦ Wavelet Transformation �Use wavelet function to approximate image. �Use with JPEG 2000, MPEG 4 Both are Lossy Compression Algorithms 26

www. hndit. com DCT vs Wavelet Algorithm Comparison Original Image (154 KB) Compress to 3 KB (1: 50) DCT Wavelet 27

www. hndit. com Image Compression: Coding and Decoding original image 262144 Bytes image encoder compressed bitstream 00111000001001101… (2428 Bytes) image decoder compression ratio (CR) = 108: 1 28