Lecture 14: Signal Compression (continued)AnnouncementsGoals for TodayFrequency Content of ImagesAnother ExampleImage vs. Spatial FrequencySlide Number 7Tricks Used in Lossy CompressionRevisiting the RoseExample: JPEGLeverage Human PerceptionBack to Image Coding -- OptionsJPG: JPEG “lossy” compressedJPEG “lossy” compressedJPEG “lossy” compressedJPEG “lossy” compressedBitmap vs JPEG “lossy” GIF “lossless” but only 256 colorsGIF 256 color vs PNG “true color”GIF 256 color “lossless” vs JPEG “lossy”GIF 256 color “lossless” vs JPEG “lossy”GIF 256 color “lossless”GIF 256 color “lossless” vs JPEG “lossy”GIF 256 color vs PNG “true color”Lossy vs. Lossless Compression?Lecture 14: Signal Compression (continued)The Digital World of MultimediaProf. Mari OstendorfEE299 Lecture 148 Feb 2008Announcements HW 4 – writing assignment only due today Labs: By tomorrow: Get your Lab2&3 sounds in CollectIt so I can showcase them Get your Lab3 *DESCRIPTION* to get full credit for the lab By Monday: submit Lab 4 exercise 1 to CollectIt Guest speaker on Monday talking about audio codingEE299 Lecture 148 Feb 2008Goals for Today EE299 Sound Showcase Review key idea of frequency in images Collaborative quiz Lossy compression Comparing types of image compressionEE299 Lecture 148 Feb 2008Frequency Content of ImagesOriginal imageFull image DCT DCT on blocksEE299 Lecture 148 Feb 2008Another ExampleEE299 Lecture 148 Feb 2008Image vs. Spatial FrequencyLPFlow 2-D frequencyhigh 2-D frequencyEE299 Lecture 148 Feb 2008Color version of block DCTOriginal imageB&W version of block DCTQUIZImage processingWhat did I do in the image processing?EE299 Lecture 148 Feb 2008Tricks Used in Lossy Compression Transform signal into a representation (basis functions) that is: More efficient (can’t reconstruct approximate version by dropping low weight components) Compatible with human perception (throw out or more coarsely quantize things humans are less sensitive to) Leverage redundancy by predicting the next sample and then quantizing the error (vs. the sample itself) Works in transform domain as wellEE299 Lecture 148 Feb 2008Revisiting the RoseEE299 Lecture 148 Feb 2008Example: JPEG Encoding/Compression Transform: Divide the image into blocks of 8x8 pixels Perform the discrete cosine transform (DCT) on each block Quantize the coefficients in each block (lossy step) Lossless compression Reorder according to increasing spatial frequency Use entropy (or arithmentic) coding on the resulting values Decoding/Decompression Undo lossless coding & reordering Reconstruct the signal Perform inverse DCT on quantized coefficients for each blocks Put the blocks back togetherEncode DecodeStorageorComm.EE299 Lecture 148 Feb 2008Leverage Human Perception Translate signal to a domain that matches perception (e.g. frequency) Image coding Small color changes are less well perceived than small changes in brightness Prominent objects distract viewer from small details We’re more sensitive to edges than background Audio coding (Orsak et al. pp. 326-328, more on Monday) We can’t hear frequencies <20Hz and >20kHz Our “quiet threshold” is higher for low and high frequencies  With simultaneous sounds close in frequency, a loud one can “mask” a soft one (i.e. we can’t hear the soft one)EE299 Lecture 148 Feb 2008Back to Image Coding -- Options bmp – bit map, no compression Lossless compression:  png (true color), gif (256 colors indexed), lossless jpg Typical compression factor around 2 Different quality factors for jpg can give compression factors ranging from 5-100EE299 Lecture 148 Feb 2008JPG: JPEG “lossy” compressedFile size = 76 KBQuality = 90EE299 Lecture 148 Feb 2008JPEG “lossy” compressedFile size = 28 KBQuality = 50EE299 Lecture 148 Feb 2008JPEG “lossy” compressedFile size = 14 KBQuality = 15EE299 Lecture 148 Feb 2008JPEG “lossy” compressedQ=75, 41 KBQ=15, 14 KB3x enlarged cropEE299 Lecture 148 Feb 2008Bitmap vs JPEG “lossy”Q=90, 76 KB1,153 KBEE299 Lecture 148 Feb 2008GIF “lossless” but only 256 colors279 KBEE299 Lecture 148 Feb 2008GIF 256 color vs PNG “true color”688 KB279 KBEE299 Lecture 148 Feb 2008GIF 256 color “lossless” vs JPEG “lossy”Q=98, 246 KB256 color, 279 KBEE299 Lecture 148 Feb 2008GIF 256 color “lossless” vs JPEG “lossy”Q=80, 77 KB256 color, 279 KBEE299 Lecture 148 Feb 2008GIF 256 color “lossless”256 color, 9 KBEE299 Lecture 148 Feb 2008GIF 256 color “lossless” vs JPEG “lossy”Q=10, 14 KB256 color, 9 KBEE299 Lecture 148 Feb 2008GIF 256 color vs PNG “true color”24bit color, 7 KB256 color, 9 KBEE299 Lecture 148 Feb 2008Lossy vs. Lossless Compression? Lossy: (jpg) Good for signals that humans perceive Good for natural images Lossless: (gif, png) Good for signal involving machine analysis Good for written/numeric documents, graphs, etc. Of course, often you use both! (as in mp3 & jpeg) What about medical signals, biometric signals, or other signals used for decision making?  Engineering and political considerations Changes in technology could change the choice of what to