mdich@luke (08/03/90)
Recently in discussing image compression several people have mentioned a technique called JPEG. Does anyone know what this is? Where can I find a description of it? Has it been adopted as any type of standard, and if so, by whom? Thanks in advance for any info. mike
mdich@schizo.samsung.com (Mike Dichiappari) (08/03/90)
In discussing various image compression schemes recently, several people have mentioned a technique called JPEG. 1. What is JPEG? 2. Who publishes the actual JPEG specs and how can I get a hold of it? 3. Has anyone actually adopted this technique as a standard? 4. I know it is a lossy technique, but what is the general consensus as to its results? Thanks in advance for any info. mike
cfogg@milton.u.washington.edu (Chad Fogg) (08/04/90)
In article <14762@samsung.samsung.com> mdich@schizo.samsung.com (Mike Dichiappari) writes: >In discussing various image compression schemes recently, several people have >mentioned a technique called JPEG. JPEG is the committe overseeing the standardization of the still frame compression algorithm. MPEG (Motion Picture Experts Group) is a sister committee handling the interframe compression algorithm. Both systems are based on Discrete Cosine Transform (DCT). I've seen a section or two devoted to DCT in a few image processing technical reference books. >1. What is JPEG? Quoting from C-Cube's CL550 background information booklet: JPEG, or Joint Photographic Experts Group, is a joint International Telegraph and Telephone Consultive Committee (CCITT) and International Standards Organization (ISO) committee in charge of defining a compression standard for continuous tone images. CCITT and ISO are responsible for the Facsimile Group III standard published in 1980. JPEG is an international open system standard. It guarantees that systems designed by different manufacturers will be able to communicate and exchange files. The proposed JPEG algorithm also guarantees high image quality at high compression ratios. Finally, the proposed JPEG algorithm is symmetrical, as it requires the same computational effort to compress or decompress an image. >2. Who publishes the actual JPEG specs and how can I get a hold of it? Electronic Imaging publishes research reports on image compression. Otherwise, one could go about the usual method of retrieving ISO/CCITT documents. >3. Has anyone actually adopted this technique as a standard? Yes, ISO/JPEG :-) There are already developers who are using either C-Cube's DCT image compression chip (described below) or the Inmos chip which, like C-Cube, claims 10:1 on an 1MB image (640*480*24bits) with no unreasonable distortion *plus* upto 40:1 interframe (yielding a net 400:1 real time video). There's even an ISDN videophone standard. >4. I know it is a lossy technique, but what is the general consensus as to its >results? > I haven't seen the results upclose yet, but what I had witnessed looked impressive (although certain areas seem to be slightly darker on the decompressed image than on the original). Intel's DVI which has been reported over the past two years to use Delta Pyramid Encoding, Laplacian Pyramid, and "Region Encoding" promises to include JPEG and MPEG compatibility. I hear CD-I now too has adopted the ISO algorithms for 72 minutes of full-frame 30fps digital interactive video on a CD-ROM, like DVI. Here are the C-Cube highlights taken from the CL550's info kit: By eliminating data redundancies image compression processors significantly reduce the data size of digital images. The attached image (Lady with Roses) has been compressed by 36:1 (97% of the date has been eliminated) but they eye can barely tell the difference between the original 16 MegaByte image, and the compressed 444 KiloByte image. The C-Cube CL550 is the first compression processor to implement the proposed JPEG standard algorithm for continuous tone image compression. It is a single-chip device capable of compressing up to 13.5 MPixels/second. The untrained eye will not detect any difference when a [1MB] screen image is compressed by a 10:1 or a [25MB] print image is compressed by 25:1 using the proposed JPEG algorithm. At 25:1 compression 96% of the data is eliminated, but the information content to which the eye is sensitive is almost entirely preserved. The CL550 compression ratio is fully programmable. In a motion sequence it can be changed on a frame by frame basis. Compression ratios as high as 200:1 are achievable. The single-chip CL550 includes a Discrete Cosine Transform and Inverse Discrete Cosine Transform unit, a Quantitizer unit, a Huffman Coder and the Quantatizer and Huffman tables necessary to implement the JPEG algorithm. The CL550 also includes a Video Bus Interface and Host Bus Interface unit that minimize the glue logic required to interface the CL550 to the pixel bus and to the system host bus. The CL550 supports 8-bit grayscale, 24-bit RGB, CMYUK, and YUV 4:4:4, 4:2:2, and 4:1:1 color spaces. A CMOS VLSI, 1.2 micron, close to 400K transistors, dissipates 1.5 Watts. The CL550 has more than 320 stages of pipeline. Each pipeline stage implements a step of the compression/decompression algorithm. Each step is equivalent to one operation. The CL550-27 is clocked at 27MHz, therefore every second the 320 stages of pipeline are clocked 27 Million times and perform close to 10 billion operations. The CL550 comes in two speed grades: CL550-10 and CL550-27. The CL550- 10 can process up to 5MPixels/second (still images), while the CL550-27 can process up to 13.5 MPixels/second (still images and motion video). Multiple CL550s can compress an image in parallel. For example, four CL550s can compress or decompress a High Definition TV (HDTV) video stream in real time.
jk87377@korppi.tut.fi (Kouhia Juhana Krister) (08/05/90)
In article <14762@samsung.samsung.com> mdich@schizo.samsung.com (Mike Dichiappari) writes: >In discussing various image compression schemes recently, several people have >mentioned a technique called JPEG. > >2. Who publishes the actual JPEG specs and how can I get a hold of it? See: Andy C. Hung Image Compression: The Emerging Standard for Color Images IEEE Computing Futures 1990, Inaugural Issue, pp. 20-29 "This article will focus on coding methods used in the proposed JPEG baseline system and parts of the JPEG extended system." Juhana Kouhia jk87377@tut.fi