BorlandTalk.com Borland discussion newsgroups   Archives   FAQ   Search   Memberlist   Usergroups   Register   Profile   Log in to check your private messages   Log in  ScaleDown Goto page 1, 2  Next          BorlandTalk.com Forum Index -> Delphi Language BASM View previous topic :: View next topic   Author Message Renate Schaaf Guest Posted: Tue Oct 11, 2005 8:36 am    Post subject: ScaleDown Hi I was wondering about the status of the ScaleDown project (scaling down a bitmap). It looks only half started, but it would be nice to get a competition going for this, as fast scaling routines come in very handy. I see several problems though, e.g. concerning validation, the signature of the routine and the actual algorithm used. Is there still interest? I'd be willing to spend some time on the project setup, if so desired. Don't know any Assembler, but I've been heavily fiddling around with bitmap operations. Renate Back to top Dennis Guest Posted: Tue Oct 11, 2005 8:44 am    Post subject: Re: ScaleDown Hi Renate Yes this challenge is in an awfull state and has been forever. I am very interested in having you volunteer to make the B&V. What does the community say? Best regards Dennis Back to top Eric Grange Guest Posted: Tue Oct 11, 2005 9:39 am    Post subject: Re: ScaleDown Quote: What does the community say? IIRC the main contention point was about how much imprecision and approximation was allowed in a valid result, with some entries being mathematically accurate, others mathematical approximations, and other being just "good looking". Basicly: different aims between being able to compute thumbnails as fast as possible and being able to provide quality downsampling. Eric Back to top Dennis Guest Posted: Tue Oct 11, 2005 9:54 am    Post subject: Re: ScaleDown Hi Eric This is why we divided the challenge into two. 1 is named Low Quality := LQ. The Other is named HQ Regards Dennis Back to top Renate Schaaf Guest Posted: Tue Oct 11, 2005 1:01 pm    Post subject: Re: ScaleDown In article <434b7ac2 (AT) newsgroups (DOT) borland.com>, Dennis wrote: Quote: What does the community say? Hi Dennis, while waiting for reaction of the community, I'd still like to make a suggestion right up front: Change the form of the function function ScaleDown(const SourceBitmap : TBitmap; ScaleFactor : Single) : TBitmap; into a procedure procedure ScaleDown(const SourceBitmap, TargetBitmap : TBitmap; ScaleFactor : Single); Reasons: 1. Current form requires the coder to create the target bitmap, user might forget to free the result. 2. User might create a bitmap, say Targetbitmap, then use code Targetbitmap:=ScaleDown(....) and not be aware that the memory/resource of the original Targetbitmap is now orphaned. 3. Current form is unsuitable for inclusion in a try .. finally - block for on the fly use of the result. 1 and 2 both happen in the BV-project as it is, so I guess the dangers described are real, even with experienced users. For other things, see reply to Eric. Renate Back to top Renate Schaaf Guest Posted: Tue Oct 11, 2005 1:01 pm    Post subject: Re: ScaleDown In article <434b8684$1 (AT) newsgroups (DOT) borland.com>, Eric Grange wrote: Quote: IIRC the main contention point was about how much imprecision and approximation was allowed in a valid result As far as I see all examples use average resampling target=sum_SourceXMin^SourceXMax sum_SourceYmin^SourceYmax w_(ij) source_(ij) (hope my typing math is understandable) with equal weights w_(ij)=1/NumSourcePixels NumSourcePixels=(SourceXMax-SourceXMin+1)*(SourceYMax-SourceYMin+1) That's a good general algorithm to start with, I'd say. If that can be sped up, then better algorithms can maybe follow. Problem is, depending on how SourceXMin, SourceXMax etc. are defined in a particular algorithm, the results can differ (among other things) by slight shifts in pixels. For example, Dennis' routine uses the same matrix size all over, whereas RMK uses differing matrix sizes, making sure to read in any source pixel only once. I find it hard to think of a metric that would notice that the two results are reasonably close. Right now, for validation, I can think of 3 possible courses of action: 1. Discuss which is the best precise algorithm to be used, then validate entries against a routine using that algorithm for exact match. (Probably the fairest, and easiest to implement once the discussion is over >:) 2. Come up with a number that measures closeness of results, allowing for slight shifts (but which average out on the total). Only thing that comes to my mind would be something like "convergence by measure": the area in which the two bitmaps differ by more than a certain amount has to be small. A recent post in Graphics suggests Fourier transforms, that might also be a way, but isn't clear to me right now. 3. Forget validation. Maybe take votes thumbsup/down for each entry. (Probably unfair) Just some ideas, as I (happily) see that my post got some reaction. Renate Back to top Dennis Guest Posted: Tue Oct 11, 2005 1:31 pm    Post subject: Re: ScaleDown Hi Renate First I few things that I think I remember. I will follow up later with more precision. We have two ScaleDown challenges. Only the LQ is present at the web. No work at all has been made on a B&V for the high quality challenge. It will be based on the LQ B&V I think we discussed properly and for a long time which function interface to use and decided to use a procedure as you suggest. My procedure is crap - please ignore it - delete it etc ;-) The one by Eric is fine and can be used as reference in both challenges. IMHO In the LQ challenge we can allow some slack. This is coded in the validation as a number each RGB component can be off. There also is a summed error number that functions must be within (Summed with sign and then scaled according to bitmap size) Best regards Dennis Back to top Eric Grange Guest Posted: Tue Oct 11, 2005 2:06 pm    Post subject: Re: ScaleDown Quote: As far as I see all examples use average resampling [...] Not all of them ;) Quote: That's a good general algorithm to start with, I'd say. That's an approximation which is decent only when downsampling a lot, so that partially overlapping source pixels contribution is negligible. The mathematically accurate version has to weight partially overlapping source pixels differently or you'll get artefacts whenever reducing to 20-99% of original size (approx, lower bound depends on image of course). If you want to downsample f.i. 640x640 to 512x512, proper weighting of source pixels is a must, equal weighting can result in atrocious image degradation if your source image isn't smooth. Quote: 1. Discuss which is the best precise algorithm to be used, then validate entries against a routine using that algorithm for exact match. (Probably the fairest, and easiest to implement once the discussion is over > Fairest yes, but definetely not the easiest to implement... this is why you commonly see source of the inaccurate "equal weighting" version (though you won't see it in PhotoShop) Quote: 2. Come up with a number that measures closeness of results Square deviation is usually accepted: lots of errors of amplitude 1 on color channels do not matter anywhere as much as a couple of misplaced high amplitude deviations, as long as the errors aren't systematic (ie. if you repeatedly a downsample, you shouldn't end up with a fully white/black/grey/saturated image). Quote: 3. Forget validation. Maybe take votes thumbsup/down for each entry. (Probably unfair) Will depend highly on source images... Also, an approximated downsampling can be made to look "good" if you introduce a sharpness filter (voluntarily or not), while a more accurate version would be softer... yet the sharpening approximated downsampler won't be suitable for use in a processing chain. Eric Back to top Eric Grange Guest Posted: Tue Oct 11, 2005 2:22 pm    Post subject: Re: ScaleDown Quote: procedure ScaleDown(const SourceBitmap, TargetBitmap : TBitmap; ScaleFactor : Single); I would actually go a step further to procedure ScaleDown(const SourceBitmap, TargetBitmap : TBitmap); and have TargetBitmap's dimension be specified by the caller. This for three extra reasons: - it allows anisotropic downsampling - it allows specifying target size accurately and unambiguously - target bitmap's allocated memory is stable if it's already a DIB Additionnally, having the routine support upsampling would be a nice convenience (esp. when resampling is anisotropic). Eric Back to top Renate Schaaf Guest Posted: Wed Oct 12, 2005 7:24 am    Post subject: Re: ScaleDown In article <434bc8f0$1 (AT) newsgroups (DOT) borland.com>, Eric Grange wrote: Quote: I would actually go a step further to procedure ScaleDown(const SourceBitmap, TargetBitmap : TBitmap); Seems good to me. Or maybe even Mike Lischke's form from the GraphicEx Stretch procedure procedure ScaleDown(NewWidth, NewHeight: integer; const Source, Target: TBitmap); It's a bit more foolproof. I'll be back for the harder things after I have seen all material, e.g. your procedure which Dennis is mentioning. Renate Back to top Renate Schaaf Guest Posted: Wed Oct 12, 2005 7:24 am    Post subject: Re: ScaleDown In article <434bbe4f$1 (AT) newsgroups (DOT) borland.com>, Dennis wrote: Quote: We have two ScaleDown challenges. Only the LQ is present at the web. The one by Eric is fine and can be used as reference in both challenges. IMHO I must be missing material then, I only have the things from the web, and I cannot find a procedure by Eric. Could maybe someone post the missing things to attachments? Renate Back to top Eric Grange Guest Posted: Wed Oct 12, 2005 7:39 am    Post subject: Re: ScaleDown Quote: your procedure which Dennis is mentioning. Couldn't find the old challenge code, so I just posted the adaptation I now use in the attachments. Eric Back to top Renate Schaaf Guest Posted: Wed Oct 12, 2005 9:17 pm    Post subject: Re: ScaleDown In article <434cbbe2$1 (AT) newsgroups (DOT) borland.com>, Eric Grange wrote: Quote: Couldn't find the old challenge code, so I just posted the adaptation I now use in the attachments. Thanks, got it. I take it, yours is a real bilinear resampling algo, with a triangle-shaped filter, not a box-resampling one, as the others I described. I can't figure out your precise algorithm right now, as the ASM is a bit of a black box for me at the moment , but maybe with time and a little help. I've made a test-application to point out the problems faced with validation, it's being uploaded to attachments. I compared results of test algorithms to the result of photoshop bilinear resampling. I've included routines from Mike Lischke's GraphicEx as a reference. Also, your routine and RMK and my own (similar), but the latter two actually lower quality box resampling. Anyway, your routine comes closest to photoshop, but still isn't identical, which in my opinion doesn't matter that much, it's a great looking algorithm and quite fast. But how to come up with a number/limit for validation is even less clear to me now. For example, GraphicExTriangle is also a (mathematically valid, I guess) bilinear resampling routine, but is seems to have shifts versus photoshop, causing the square distance to shoot up, a lot maybe, depending on the picture. LQ=Box-Resampling HQ=Bilinear and better? What about the real low-quality Delete_Scans, used by TImage.stretch=true? That also comes in handy when time is really pressing, and can be made faster, too. Enough rambling, Good night Renate Back to top Renate Schaaf Guest Posted: Wed Oct 12, 2005 9:56 pm    Post subject: Re: ScaleDown Sorry, I couldn't upload the project, too large, need to change the size of bitmaps included. Renate Back to top Renate Schaaf Guest Posted: Wed Oct 12, 2005 10:35 pm    Post subject: Re: ScaleDown Now in attachments. Renate Back to top Display posts from previous: All Posts1 Day7 Days2 Weeks1 Month3 Months6 Months1 Year Oldest FirstNewest First         BorlandTalk.com Forum Index -> Delphi Language BASM All times are GMT Goto page 1, 2  Next Page 1 of 2   Jump to: Select a forum C++ Builder Newsgroups----------------C++ BuilderC++ Builder (ActiveX)C++ Builder (Announcements)C++ Builder (Command Line Tools)C++ Builder DatabasesC++ Builder Databases (ADO)C++ Builder Databases (Desktop)C++ Builder Databases (InterBase Express)C++ Builder Databases (Multi-Tier)C++ Builder Databases (SQL Servers)C++ Builder (Graphics)C++ Builder (IDE)C++ Builder (Internet)C++ Builder (Internet Socket)C++ Builder (Internet Web)C++ Builder (Jobs)C++ Builder (Language)C++ Builder (Language C++)C++ Builder (Compatibility with MFC)C++ Builder (Native API)C++ Builder (Non-Technical)C++ Builder (Students)C++ Builder (Third Party Tools)C++ Builder (Upgrade)C++ Builder (VCL)C++ Builder (VCL Components Usage)C++ Builder (VCL Components Development)C++ Builder (VCL Usage)C++ Builder (WinAPI)C++ Builder Installation Delphi Newsgroups----------------Delphi (General)Delphi ActiveX Controls UsageDelphi ActiveX Controls DevelopmentDelphi As400Delphi DatabasesDelphi Databases (ADO)Delphi Databases (dbExpress)Delphi Databases (Desktop)Delphi Databases (InterBase Express)Delphi Databases (Multi-Tier)Delphi Databases (SQL Servers)Delphi DeploymentDelphi GraphicsDelphi IDEDelphi InternationalizationDelphi InternetDelphi Internet Isapi-WebBrokerDelphi Internet WebSnapDelphi Internet WinsockDelphi JobsDelphi Language BASMDelphi Language ObjectPascalDelphi Model Driven Architecture (General)Delphi Model Driven Architecture (Third Party)Delphi MultimediaDelphi Native APIDelphi Non-TechnicalDelphi ObjectPascalDelphi OLE AutomationDelphi OO designDelphi OpenToolsAPIDelphi Reporting-ChartingDelphi RTLDelphi StudentsDelphi TeamSourceDelphi Thirdparty ToolsDelphi Thirdparty Tools (General)Delphi Thirdparty Tools (IntraWeb)Delphi Thirdparty Tools (RaveReports)Delphi UpgradeDelphi VCLDelphi VCL ComponentsDelphi VCL Components UsingDelphi VCL Components WritingDelphi WebServices SOAPDelphi WinAPIDelphi XMLDelphi Installationcomp.lang.pascal.delphi.components.misccomp.lang.pascal.delphi.components.usagecomp.lang.pascal.delphi.components.writingcomp.lang.pascal.delphi.databasescomp.lang.pascal.delphi.misccomp.sources.delphi JBuilder Newsgroups----------------JBuilderJBuilder AnnouncementsJBuilder Applet IssuesJBuilder As400JBuilder CompilerJBuilder CORBA/RMIJBuilder DatabasesJBuilder DB SwingJBuilder DebuggerJBuilder DeploymentJBuilder DocumentationJBuilder EnterpriseJBuilder HandheldJBuilder IDEJBuilder JavaJBuilder JavaAPIJBuilder Java LanguageJBuilder Java SwingJBuilder Java BeansJBuilder Java Beans UsageJBuilder Java Beans DevelopmentJBuilder JBCLJBuilder JDataStoreJBuilder JobsJBuilder LinuxJBuilder MacintoshJBuilder & MobileSetJBuilder MultiLingual AppsJBuilder Non-TechnicalJBuilder OpenToolsJBuilder Servlets (JSP)JBuilder StudentsJBuilder Team DevelopmentJBuilder Thirdparty ToolsJBuilder UpgradeJBuilder Installation InterBase Newsgroups----------------InterBaseInterBase Bug DiscussionInterBase GeneralInterBase IBConsoleInterBase OpenSourceInterBase ReplicationInterBase SQLInterBase ThirdpartyInterBase Installation Miscellaneous Borland Newsgroups----------------AppCenterAppServerEnteraVisiBroker  You cannot post new topics in this forum You cannot reply to topics in this forum You cannot edit your posts in this forum You cannot delete your posts in this forum You cannot vote in polls in this forum Powered by phpBB © 2001, 2006 phpBB Group SEO toolkit © 2004-2006 webmedic.