|
Welcome to a brand of
Mathematical
Services. 2026-02-14 13:09:39 - Paul D. Foy - ImageReconstruction I think I'm continuing to view (most of) the developments in AI positively, not least in so far as the fact that they are enabling in the way I wanted to operate (but was frustrated in doing so) during my working life. I got my first (permanant) job as a research officer, where I was predominantly engaged in software development. Indeed all of my subsequent career was as a software engineer. But I was often frustrated in this role because generally Companies recruited someone to do the mathematical (or engineering) bits and they recruited a seperate person to code the results, the algorithms up. But I got a degree, and PhD as a mathematician, and I was perfectly good at maths (at least I thought so) and wanted to do the development of the methods and the algorithms AND code it up. But this was often denied or sneeered upon. I actually thought I was better at doing the design of the algorithm anyway, but there always seemed to be someone (at least whreh I was contracting) that did this. But now with AI, chatGPT it is even easier to do both jobs. The coding is a less specialised task, so you can develop the algorithm (there's a lot of help with this as well) and code it up with less trouble. Fits my view of the world, so that's fine. Of course there's going to be people that are not happy about what AI is doing, and its getting used in some pointless areas (people are using AI to do the pointless things they were doing already - I don't know how that is going to work! :)) - seems to me AI is not a danger to this they'll just do the pointless things quicker - everybody gains. 2026-02-27 05:59:54 - Paul D. Foy - It is unlikely that a . pf3 file will offer any benefits. The overhead of storing the data for 3 channels is too large to offset any benefit from processing a smaller number of entities. Tke . pf4(1) remains my contribution. 2026-02-26 07:01:13 - Paul D. Foy - What does a person do in this looming world of AI - not do anything because the power of computing is going to become so large that software engineers and algorithms will be done for you. So there is no need to bother. I still understand that AI can't do new things, can't think like a person. So as far as I understant it a . pf3 hasn't been done before, and I'm curious as to wether it will be equivalent in storage to . jpg but better quality. The British . jpg. Anyway where's the fun in not bothering! :). 2026-02-25 10:03:23 - Paul D. Foy - Each 'hue' doesn't need to be stored as an 'index' - just store the value - so 16 values to store per channel. This is why I panicked and deleted the comment. 2026-02-25 09:52:24 - Paul D. Foy - Each 'hue' doesn't need to be stored as an 'index' - just store the value - so 16 values to store per channel. This is why I panicked and deleted the comment. 2026-02-25 09:47:54 - Paul D. Foy - I got the comment again - because on this site each and every comment is preserved -= there is no hiding or faking like in real life. So whether the comment stands the test of time it is still here, and I give it to you again: I posted and advertised about the . pf4 method for lossless storage of an image (a bitmap) which is midway in size between a . jpg and . bmp (but better quality than the . jpg). It didn't seem to garner much interest. However, if I may say so, this is a lost opportunity. Unfortunately, the design, whilist certainly serving a purpose, can be improved upon. Instead of representing each colour, each RGB value, as an index (with up to 256 index values) I think a better approach would be to store each hue of each RGB channel as an index and push the . pf4 algorithm through each channel separatley - I will call this file a . pf3 file. Remember the goal is lossless storage of the . bmp with reduced storage size. My opinion, is that there has been needless complexity on the . jpg format (all manner of things seem to be involved in its encoding/decoding), whilst the underlying . pf4,3 specification is extremely simple. In fact I think some artistry is needed, or some developments on producing fewer colours such that an RGB (with up to 256*256*256 colours) can be represented, indistinguishable for the human eye, with far fewer colours. My guess is that 16 per channel (i. e. per hue of R, G, B) would do. The . pf3 algorithm on these 16 colours would be extremely efficient and done per channel I think the storage would be comparable with a . jpg. There's also interesting work on making colour equivalences of . jpgs I think. I'll get on with the . pf3 in due course, although it wont be too quick. There's not so much help from AI with new algorithms (it has to leverage what has gone before), although it can help with the rusty basics! It's also coming up to allotment time (food growing). 2026-02-25 09:42:36 - Paul D. Foy - I mistakenly deleted my earlier comment which I made this morning. I hope you got it. Post a comment: |