As a technical member of the JPEG2000 body, I have to say that sitting in and participating in the meetings was quite an interesting experience. As always, there were a number of technical issues to resolve, but the most heated and often most time was spent on IP issues. The comment in the article, warning about people holding IP crawling out of the woodwork if the standard is successful is definitely one to watch. While I have not been involved in the last couple of meetings, I'm still actively involved in following what has been happening technology wise. The benefit of wavelet compression is not so much in compression quality, but in features. From a file, one can get multiple resolutions, multiple quality levels, selective decoding of a specific region (random access), etc. This should be a benefit in the long run. The article touches on it briefly, but one down side to the new standard is that it is more resoure intensive than the current JPEG even PNG formats -- in memory and cpu power. I suspect it will be a while until we see wireless devices with the resources to handle the format in a general purpose manner. A custom ASIC solution is a possibility, but would a device like a Palm Pilot include one? What are the factors that lead to the improved quality and additional complexity? JPEG2000 is based upon wavelet compression. The standard allows for tiling images, but it is more typical that the whole image is compressed, for most applications. Unlike JPEG, where pixels are grouped into 8x8 blocks, wavelet schemes which operate upon the whole image have a lot more data to play with when it comes to throwing away (quantizing) information. If one were to attempt to use the standard with 8x8 tiles, well, it would not work very well. Because of this increase in data that is part of the working set, the amount of memory needed, when compared with JPEG is, in most cases, much greater. During the JPEG2000 process, a few companies proposed block-based wavelet solutions which would reduce this complexity without sacrificing feature set or even quality. However, they were not included due to concerns over the companies not making the IP available on a free and non-discriminatory basis. There has been a part-2 to JPEG2000 proposed that will allow for the inclusion of technologies which IP is not necessarily free. My question to the group here is, if JPEG2000 takes off and companies and individuals who have not previously declared IP come forward and want royalties, will the standard be hurt? Is there a place for a part-2 which contains IP which is not free? And, what applications does the community here see as being crucial for the adoption of JPEG2000? Good night and enjoy.. -jim
From having been involved in the Jpeg2000 work from the technical side of things, I feel safe in saying that claims of 200:1 compression is...well... nice and everything in a specialized case. In reality, the compression results for what was finally adopted as Jpeg2000 Part 1 are only about 10-20% better than JPEG baseline for the same realitive quality level -- be it error numbers, etc -- on average. At high bit rates, there is relatively little that differentiates wavelet scemes from dct based schemes. Where wavelets show better quality is at high-compression ratios (even like the example given in the EETimes article) the reason there is that the wavelet transform operates on the whole image rather than individual blocks, like the 8x8 blocks in JPEG. Thus, you have more flexibility in the data you throw away -- better energy compaction, etc. Where JPEG 2000 was supposed to be useful, according to those in charge, was not in compression performance but features. It should be noted, in addition to having more features, Jpeg2000 is also more complex than JPEG. More memory is required as well as more computational complexity. It is a tradeoff, to gain features. Will it succeed? Who knows. I think that makes some sense...anyway...
Slashdot Mirror
As a technical member of the JPEG2000 body, I have to say that sitting in and participating in the meetings was quite an interesting experience. As always, there were a number of technical issues to resolve, but the most heated and often most time was spent on IP issues. The comment in the article, warning about people holding IP crawling out of the woodwork if the standard is successful is definitely one to watch. While I have not been involved in the last couple of meetings, I'm still actively involved in following what has been happening technology wise. The benefit of wavelet compression is not so much in compression quality, but in features. From a file, one can get multiple resolutions, multiple quality levels, selective decoding of a specific region (random access), etc. This should be a benefit in the long run. The article touches on it briefly, but one down side to the new standard is that it is more resoure intensive than the current JPEG even PNG formats -- in memory and cpu power. I suspect it will be a while until we see wireless devices with the resources to handle the format in a general purpose manner. A custom ASIC solution is a possibility, but would a device like a Palm Pilot include one? What are the factors that lead to the improved quality and additional complexity? JPEG2000 is based upon wavelet compression. The standard allows for tiling images, but it is more typical that the whole image is compressed, for most applications. Unlike JPEG, where pixels are grouped into 8x8 blocks, wavelet schemes which operate upon the whole image have a lot more data to play with when it comes to throwing away (quantizing) information. If one were to attempt to use the standard with 8x8 tiles, well, it would not work very well. Because of this increase in data that is part of the working set, the amount of memory needed, when compared with JPEG is, in most cases, much greater. During the JPEG2000 process, a few companies proposed block-based wavelet solutions which would reduce this complexity without sacrificing feature set or even quality. However, they were not included due to concerns over the companies not making the IP available on a free and non-discriminatory basis. There has been a part-2 to JPEG2000 proposed that will allow for the inclusion of technologies which IP is not necessarily free. My question to the group here is, if JPEG2000 takes off and companies and individuals who have not previously declared IP come forward and want royalties, will the standard be hurt? Is there a place for a part-2 which contains IP which is not free? And, what applications does the community here see as being crucial for the adoption of JPEG2000? Good night and enjoy.. -jim
From having been involved in the Jpeg2000 work from the technical side of things, I feel safe in saying that claims of 200:1 compression is ...well... nice and everything in a specialized case. In reality, the compression results for what was finally adopted as Jpeg2000 Part 1 are only about 10-20% better than JPEG baseline for the same realitive quality level -- be it error numbers, etc -- on average. At high bit rates, there is relatively little that differentiates wavelet scemes from dct based schemes. Where wavelets show better quality is at high-compression ratios (even like the example given in the EETimes article) the reason there is that the wavelet transform operates on the whole image rather than individual blocks, like the 8x8 blocks in JPEG. Thus, you have more flexibility in the data you throw away -- better energy compaction, etc. Where JPEG 2000 was supposed to be useful, according to those in charge, was not in compression performance but features. It should be noted, in addition to having more features, Jpeg2000 is also more complex than JPEG. More memory is required as well as more computational complexity. It is a tradeoff, to gain features. Will it succeed? Who knows. I think that makes some sense...anyway...