StuffIt Image Compression Whitepaper

StuffIt® Image Compression White Paper Date: 1/5/2006 (Revision 2.1) The technology that lies behind StuffIt Image represents an important breakthrough in the field of data compression. This breakthrough allows StuffIt to re-compress JPEG files by an average of nearly 30% without causing any addition loss in image quality. In the past JPEG’s have been notoriously resistant to existing lossless compression methods, such as Zip, as they already contain a lossless compression component as part of their file format. The patents filed for StuffIt image cover a new method for preprocessing JPEG images before they are compressed, and it the nature of this preprocessing that explains why StuffIt is able to yield high compression ratios on files that Zip is typically not able to compress at all. StuffIt Image also includes technology that allows the lossless compression of 24-bit images. Using StuffIt’s 24-bit lossless compression method, it is now possible to store full quality images within a file size that is equivalent to a high quality JPEG. This new technology was initially introduced as part of StuffIt Deluxe® 9, Smith Micro’s award winning archive and compression software program. StuffIt Deluxe allows PC users to reduce the size of JPEG images by placing them inside StuffIt X archives. While this approach is ideal for computer users looking to save space and speed up Internet file transfers, Smith Micro also offers its image compression technology outside the context of StuffIt X archives to better enable its implementation in an embedded environment. The same technology that enables consumers and businesses to compress JPEGs on their PC can also be used to increase storage and bandwidth efficiency in a wide variety of portable multi-media devices such as wireless handsets, cameras, and PDA’s. Whether on the Desktop, in an Enterprise environment, or on the go, StuffIt’s image compression technology provides an important new tool for those looking to reduce storage and bandwidth costs. Key Talking Points • JPEG Compression: Compress JPEG files an additional 20 to 30 percent without causing additional loss in image quality. • 24bit Image Compression: StuffIt provides a lossless compression method that compresses 24bit full quality images up-to 70%. 1-15 StuffIt Image White PaperJPEG Compression The image compression technology provided by StuffIt Image Compression builds on nearly 20 years of experience in the field of data compression. StuffIt’s legacy is one of continuous innovation; first released in 1986, StuffIt technology has continued to evolve to provide solutions to some of the most common problems faced by those in the data compression industry. The ability to further compress previously “compression resistant” JPEG files is the latest such development. Traditionally, JPEG images have been classified as “un-compressable” because the JPEG file format already contains a lossless compression component. The output of this compression component has appeared to other compression programs as a random/uncompreessibl string of bits. Figure 1: Creating JPEG As illustrated by Figure 1, the creation of a JPEG image can be split into two basic functions. The first function covers the generation of loss. The original image is first down-sampled to reduce the overall image quality. Then a Discrete Cosine Transform (DCT) is applied. The second function takes this data and applies a lossless compression algorithm (Huffman) to further reduce the file size. It is this second function that is the key to understanding the nature of StuffIt’s compression breakthrough. Once the Huffman encoding layer has been applied, other lossless compression methods are typically unable to further reduce the size of the file. When a JPEG is compressed with a standard Windows compression method, such as zip, there is typically little or no reduction in file size. The random nature of the Huffmancomprresse data provides little or no structure for typical compression methods to analyze. StuffIt’s Image compressor does not attempt to compress the near-random data output of a JPEG file, instead it seeks to convert this data to a state more conducive to compression. 2-15 StuffIt Image White PaperFigure 2: StuffIt JPEG Compression As illustrated by Figure 2, this conversion involves decompressing the Huffmancomprresse image blocks and then re-compressing them with a new method specifically designed to handle the data produced by the first part of the JPEG process. Any header data (such as EXIF metadata) is separately compressed with a generic compressor and stored with the file. When a compressed JPEG is decompressed, this process is reversed; the StuffIt compression method is removed, and the original Huffmma encoding is re-applied to the file. StuffIt achieves its 20 to 30 percent compression gain by replacing an outdated compression scheme with one custom-designed for the task at hand. Again, it is important to note that the StuffIt compression used is lossless. There is no reduction in image quality. When a JPEG file that has been compressed with StuffIt is expanded and compared to the original file, it can be seen as an exact replica at both the pixel level and the byte level. The true lossless nature of StuffIt’s compression method provides the key differentiator between StuffIt’s image compression technology, and other image compression methods such as JPEG 2000. While the process of simply opening and re-saving a JPEG image results in image information being lost, images can be converted back and forth between StuffIt and JPEG format without this loss occurring. While StuffIt Image Compression technology allows for the compression of existing JPEG files, it will also provide manufactures with the ability to bypass the JPEG format and write out images directly in StuffIt format. For this “Bitmap” level interface, StuffIt will acquire sensor data directly from the camera chipset. 3-15 StuffIt Image White PaperFigure 3: JPEG Compression from Bitmap When the bitmap interface technique is used (Figure 3), the captured image is never written out in JPEG format. The loss-generating portion of the JPEG creation process is still employed, but it is handled by StuffIt, rather than a legacy JPEG implementation. After the DCT data has been created, StuffIt’s lossless encoding method is applied directly instead of applying Huffman encoding (which would result in a standard JPEG file). The fact that it is not necessary to first decompress Huffman-encoded image blocks when using the Bitmap interface provides a mechanism through which images can be saved in StuffIt format more efficiently. Images saved in this format can be converted back to a JPEG file simple by decompressing the image blocks compressed with StuffIt’s lossless method, and re-compressing them with Huffman encoding. 4-15 StuffIt Image White PaperJPEG Image Compression Results Typically, users will be able to reduce the size of a JPEG image anywhere from twenty to thirty percent. As with all file compression technologies, individual results may vary. Some files will provide results significantly better than thirty percent while others will compress less than twenty percent. In order to provide an objective test set, two image sets have been used. One sample set utilizes a default set of Kodak reference images that are commonly used in the image compression industry to demonstrate the effectiveness of various methods, the other sample set is drawn from a FujiFilm FinePix F10 6.3 MegaPixel Camera. The performance results generated for this white paper were conducted on an ARM 9 platform with a 266mhz processor. The original Kodak images are un-compressed .png files; these were then converted to JPEG’s on a Macintosh using GraphicConverter X by Lemke Software GmbH. A value of 40 (on a scale of 1 – 100, where 1 is the worst quality and 100 the best) was used to specify JPEG quality. The images for the FujiFilm F10 were taken using the maximum available resolution. The images used to perform the test can be downloaded from http://www.stuffit.com/imagecompression.html 5-15 StuffIt Image White PaperKodak References Images Kodim01.jpg Kodim02.jpg Kodim03.jpg Kodim04.jpg Kodim05 Kodim06 Kodom07 Kodim08 Kodim09 Kodim10 Kodim11 Kodim12 Kodim13 Kodim14 Kodim15 Kodim16 Kodim17 Kodim18 Kodim19 Kodim20 Kodim21 Kodim22 Kodim23 Kodim24 Table 1: Kodak Reference Images In order to generate the numbers presented in Table 2 (shown below), the files described in Table 1 were compressed using a version of StuffIt Image running on a AMD64 3200 based desktop PC running Windows XP Professional. “Time to Compress” describes the time required to take the JPEG file and compress it into StuffIt format. “Time to Decompress” refers to the time required to take the StuffIt-compressed version of that file and convert it back to JPEG. 6-15 StuffIt Image White PaperFile Size (in bytes) Compressed Size Saved Time to Compress (in seconds) Time to Decompress Kodim01.jpg 53,801 40,258 25% 0.11 0.11 Kodim02.jpg 28,536 19,577 31% 0.08 0.06 Kodim03.jpg 26,150 17,199 34% 0.06 0.05 Kodim04.jpg 31,778 21,994 30% 0.06 0.08 Kodim05.jpg 60,307 46,127 23% 0.13 0.11 Kodim06.jpg 43,340 31,660 26% 0.09 0.09 Kodim07.jpg 32,741 22,249 32% 0.06 0.06 Kodim08.jpg 60,405 45,259 25% 0.11 0.13 Kodim09.jpg 26,983 17,494 35% 0.06 0.05 Kodim10.jpg 29,717 20,185 32% 0.06 0.06 Kodim11.jpg 39,613 28,282 28% 0.09 0.08 Kodim12.jpg 27,950 18,141 35% 0.06 0.06 Kodim13.jpg 65,083 51,882 20% 0.14 0.13 Kodim14.jpg 48,842 36,489 25% 0.09 0.09 Kodim15.jpg 29,260 19,644 32% 0.06 0.06 Kodim16.jpg 32,927 22,757 30% 0.06 0.08 Kodim17.jpg 32,766 22,544 31% 0.06 0.08 Kodim18.jpg 46,969 36,117 23% 0.11 0.09 Kodim19.jpg 36,787 25,990 29% 0.08 0.08 Kodim20.jpg 26,843 17,768 33% 0.06 0.06 Kodim21.jpg 37,466 27,494 26% 0.09 0.06 Kodim22.jpg 37,694 27,585 26% 0.08 0.08 Kodim23.jpg 24,223 15,306 36% 0.06 0.06 Kodim24.jpg 46,517 35,578 23% 0.09 0.09 Average: 29% 0.08 0.08 Table 2: JPEG Compression Performance for Kodak Reference Image Set Table 2 shows that across the range of images provided in the Kodak data set, the compression ratio achieved remains in the 20 to 30 percent range. The average time for both compression and decompression phases is 0.08 seconds. 7-15 StuffIt Image White PaperFujiFilm FinePix F10 Images FUJI01 FUJI02 FUJI03 FUJI04 FUJI05 FUJI06 FUJI07 FUJI08 FUJI09 FUJI10 FUJI11 FUJI12 FUJI13 FUJI14 FUJI15 FUJI16 FUJI17 FUJI18 FUJI19 FUJI20 Table 3: FujiFilm F10 Images In order to generate the numbers presented in Table 4 (shown below), the files described in Table 3 were compressed using a version of StuffIt Image running on a AMD64 3200 based desktop PC running Windows XP Professional. “Time to Compress” describes the time required to take the JPEG file and compress it into StuffIt format. “Time to Decompress” refers to the time required to take the StuffIt-compressed version of that file and convert it back to JPEG. 8-15 StuffIt Image White PaperName Size (in bytes) Compressed Size Ratio Time to Compress (in seconds) Time to Decompress FUJI01.JPG 3,113,986 2,345,046 25% 5.11 4.91 FUJI02.JPG 3,071,310 2,218,681 28% 5.06 4.67 FUJI03.JPG 3,002,523 2,270,604 25% 5.42 5.23 FUJI04.JPG 2,882,860 2,159,791 26% 5.94 5.09 FUJI05.JPG 3,021,845 2,245,671 26% 6.02 5.27 FUJI06.JPG 2,950,315 2,256,528 24% 5.66 5.83 FUJI07.JPG 2,930,731 2,161,435 27% 5.27 4.86 FUJI08.JPG 2,991,176 2,297,733 24% 6.03 5.61 FUJI09.JPG 3,193,463 2,397,716 25% 5.45 4.94 FUJI10.JPG 2,984,577 2,302,179 23% 5.81 5.36 FUJI11.JPG 2,998,293 2,262,757 25% 5.80 5.25 FUJI12.JPG 3,021,714 2,230,045 27% 6.16 5.75 FUJI13.JPG 3,072,758 2,291,947 26% 5.66 4.97 FUJI14.JPG 3,120,710 2,339,260 26% 5.34 5.31 FUJI15.JPG 3,028,222 2,353,621 23% 5.41 4.94 FUJI16.JPG 3,244,984 2,475,617 24% 5.73 5.25 FUJI17.JPG 2,859,975 2,079,397 28% 5.08 4.72 FUJI18.JPG 2,939,887 2,253,994 24% 6.03 5.63 FUJI19.JPG 3,058,115 2,305,175 25% 5.88 5.25 FUJI20.JPG 3,259,021 2,436,885 26% 5.73 5.16 Average 25.35% 5.63 5.20 Table 4: JPEG Performance for FujiFilm FinePix F10 Image Set Table 4 illustrates that across the range of images included in the FujiFilm FinePix F10 image set, the average compression ration achieved is 25%. The average time to both compress and decompress is a little over 5 seconds. 9-15 StuffIt Image White PaperCompression Ratio Variation by JPEG Level One factor known to have a significant impact on compression efficiency is the quality level of the JPEG being compressed. In the context of JPEG images, image quality measures the amount of loss generated from the original image. The example in table 5, using Kodim23 from the Kodak reference image set, illustrates the extremes. The first image is compressed with JPEG level 100, the highest quality level, while image 2 is compressed at level 0, the lowest available setting. JPEG Level 100 (271,887 bytes) JPEG Level 0 (7,8200 bytes) Table 5: Kodim23 when converted to JPEG level 100 and JPEG level 0 As the performance of StuffIt’s JPEG compression method is measured across the range of JPEG quality settings it can be seen that the compression ratio declines as the amount of image information increases. The information presented in Tables 6 and 7 illustrates the way in which the compression ratio changes with the JPEG quality level. 10-15 StuffIt Image White PaperJPEG Level Size Compressed Size Ratio 100 271,887 216,873 21% 90 77,329 56,538 27% 80 48,757 34,508 30% 70 37,812 25,989 32% 60 31,631 21,125 34% 50 27,754 18,111 35% 40 24,223 15,306 37% 30 20,620 12,543 40% 20 16,427 9,196 45% 10 11,638 5,303 55% 0 7,820 1,822 77% Table 6: Compression Ratios for Kodim23 at various JPEG quality levels Table 7: A line chart illustrating the compression ratio for Kodim23 at various JPEG Quality levels 0% 10% 20% 30% 40% 50% 60% 70% 80% 100 90 80 70 60 50 40 30 20 10 0 11-15 StuffIt Image White PaperLossless Compression of 24-bit Images In both image compression scenarios described above, the common factor is that picture quality is lost during the JPEG down-sample and DCT phases. While the ability to further compress JPEG images without imposing additional quality loss is an important attribute of StuffIt’s JPEG compression, it remains the case that the resulting image is a compromised version of the bitmap that was originally produced by the camera sensor. In addition to the compression of JPEGs, StuffIt’s image compression technology also encompasses a 24-bit lossless compression method that by-passes JPEG completely. When compressing full-quality 24-bit images, StuffIt includes specialized modes for handling both “natural” images (such as photographs) and "artificial" images (such as computer generated graphics). StuffIt’s ability to analyze data as it is being added to an archive provides a key advantage; StuffIt is able to automatically select the most efficient compression method for the type of data being provided, and is even able to make a determination, when analyzing an un-compressed 24-bit picture, whether the artificial or natural algorithm should be used. Significantly, the lossless compression of full quality images that StuffIt Image Compression can provide will allow users to store full resolution images within the similar resulting size as a high quality JPEG, but with no loss of image information. The following table illustrates the size performance of StuffIt’s lossless bitmap compression method compared to other lossless formats such as TIFF and PNG. As with the previous performance data, the Kodak sample image set was used. The numbers presented in Table 4 were generated by saving the files described in Table 1 in a number of lossless formats; the formats used for the test were Bitmap (.bmp), Portable Network Graphic (.png), StuffIt Image Format (StuffIt, or .sif), and Tag Image File Format (.tif). 12-15 StuffIt Image White PaperFile BitMap (.bmp) Size in bytes StuffIt (.sif) size in bytes PNG size in bytes TIFF size in bytes Kodim01 1,179,702 395,732 736,501 847,605 Kodim02 1,179,702 321,394 617,995 726,568 Kodim03 1,179,702 278,725 502,888 578,584 Kodim04 1,179,702 344,155 637,432 741,834 Kodim05 1,179,702 446,769 785,610 917,568 Kodim06 1,179,702 373,849 618,959 703,938 Kodim07 1,179,702 304,547 566,322 642,958 Kodim08 1,179,702 304,547 788,470 981,245 Kodim09 1,179,702 316,997 582,899 659,494 Kodim10 1,179,702 328,366 593,463 685,355 Kodim11 1,179,702 358,035 621,023 716,280 Kodim12 1,179,702 305,127 531,024 607,782 Kodim13 1,179,702 491,848 822,712 973,990 Kodim14 1,179,702 411,739 692,201 793,285 Kodim15 1,179,702 331,895 612,582 730,701 Kodim16 1,179,702 312,594 534,247 605,267 Kodim17 1,179,702 324,963 602,078 680,780 Kodim18 1,179,702 450,052 780,947 914,233 Kodim19 1,179,702 370,363 671,476 767,113 Kodim20 1,179,702 294,238 492,462 552,468 Kodim21 1,179,702 372,878 637,051 725,661 Kodim22 1,179,702 414,250 701,970 836,188 Kodim23 1,179,702 335,196 557,596 665,477 Kodim24 1,179,702 391,132 706,397 827,587 Total Image Data 28,312,848 8,579,391 15,394,305 17,881,961 Table 8: StuffIt Lossless Compression vs PNG and TIFF (Size in bytes) Table 8 illustrates the performance of StuffIt’s lossless compression method against PNG and TIFF. The Kodak data set, described above, was used to generate these results. 13-15 StuffIt Image White PaperTable 9: Total Image Data Comparison Between StuffIt, BMP, TIFF, and PNG Table 9 illustrates a comparison between the total number of bytes required to store the image data for all the pictures in Kodak image set with no loss. StuffIt’s lossless compression offers approximately a 45% advantage compared to the size of the PNG, the next best format. Support for Additional File Types As the multimedia capabilities of handsets increases, StuffIt will offer solutions designed to provide the same benefits currently available for images to movies, sounds, and other rich media content. The techniques used to convert JPEG images into a compressible format can also be applied to other media file formats where a lossless compression component is present. In the case of MP3 audio files, StuffIt will be able to achieve a file size reduction of 15% (on average), and in the case of MPEG movie files it is anticipated that a saving of 15%– 25% can be realized. BMP TIFF PNG SIF 0 7,500,000 15,000,000 22,500,000 30,000,000 Lossless Image Formats 8,579,391 15,394,300 17,881,960 28,312,850 14-15 StuffIt Image White PaperContact Information For more information about StuffIt Image contact: W. Rick Wyand VP of Sales -Wireless and OEM Smith Micro Software, Inc. 51 Columbia Aliso Viejo, CA 92656 949-362-2347 -Office 949-362-5723-Fax 913-220-5244-Mobile rwyand@smithmicro.com 15-15 StuffIt Image White Paper