Thank you for your comments and I'm sorry for my response has been delayed. As you've know, my software is an application of the inverse matrix solution method which is explained in the following site; http://139.134.5.123/tiddler2/c22508/iteration.htm As the review, when applying Fourier transform to deconvolve the blurred image, firstly the output image g(x) is formulated with the true image f(x) as, g(x) =3D f*n(x) where the extent of n(x) corresponds to the deviation from the focal point (if g(x) is just in focus, then n(x) =3D ((delta))(x).) After applying Fourier transform to the above equation, we get; G(u) =3D F(u)N(u) hence, F(u) =3D G(u)/N(u) Therefore, by applying inverse Fourier transform to G(u)/N(u), we can get the image data f(x) before convolution. While my method is to solve the first equation: g(x) =3D f*n(x) directly. Firstly that equation is discretized as, g(j) =3D ((sigma))_k f(j-k)n(k) It is expressed with matrix-vector form as, g =3D Af hence, f =3D A^(-1)g The problem is how to calculate f(x) from the above equation. You may imagine to apply any standard iterative inverse matrix solution methods such as Gauss-Seidel method for it. But we must be careful to apply Gauss-Seidel method to this specific problem because A doesn't satisfy the condition which Gauss-Seidel method requires, while my iterative inverse matrix solution method is enough robust to deal with that problem (the convergent speed is much slower, though.) jim wrote:> Jerry Avins wrote: > > > > > You're inferring what OP must have done from your knowledge of what > > can't work. > > Well excuse me, I wasn't aware using knowledge was not permitted. I was > also using the information that his images supplied. After all, if "it > is nothing but a 2D FIR filter", as Rune put it, the images themselves > explain everything. > > >I'm trying to understand what he did from what he said he > > did. I agree that what he said he did can't work, > > No, you are trying to prove your interpretation of what is written on > the software web page can't work. > > > > and I'l like a clarification from him. > > If you are so desperate to know what the impulse response is for the > processing at each of the different parameter settings available why not > download the program and feed it an impulse and see what the response is > for each setting. > > >Tit you catch Rune's observation that this > > thread has only half of his remarks? > > The OP wrote in a different thread: > > > The pictures shown in the following URL; > http://www.unmannedspaceflight.com/index.php?act=3DAttach&type=3Dpost&id==3D2017> are the original image of Vesta (left), its de-convoluted image (middle > / focus depth:=3D3.8, iteration:=3D8 / software: Focus Corrector ) and > another de-convoluted image (right / focus depth:=3D4.3, iteration:=3D8 / > software: Focus Corrector). > > > With his statement "focus depth:=3D4.3, iteration:=3D8" that is enough ri=ght> there to know he is not applying a simple kernel you can write down with > 3 lines of ascii. So I don't know why you continue to beat that dead > horse. > > -jim > > > > > > Even if you missed the part about there being input parameters =and> > > thought he was using a single filter once only you should know it's n=ot> > > that filter that produced those images. Simply by looking at the > > > pictures you can tell he is using something better than that. That > > > filter would leave some blocky artifacts that would be noticeable. > > > Running that filter over and over would simply magnify the blockiness. > > > So your interpretation is clearly wrong. According to Rune that makes > > > you a charlatan because that's what charlatans do - fail to explain h=ow> > > something works. > > > > Iterating with that filter is equivalent to a single pass with a larger > > filter, it that's what you mean. > > > > > I assume he posted here to get people to download and try the p=rogram> > > and offer criticism based on what they find using their own images. I > > > don't think he was looking for critiques on what people would imagine > > > they might find. > > > > I'm sure you're right about that. If I get a chance, I'll do that and > > have a go at the three-image comparison I described. > > > > Jerry > > -- > > Engineering is the art of making what you want from things you can get. > > =AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF==AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF= =AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF> > ----=3D Posted via Newsfeeds.Com - Unlimited-Unrestricted-Secure Usenet N=ews=3D----> http://www.newsfeeds.com The #1 Newsgroup Service in the World! 120,000+ =Newsgroups> ----
Unmasked Tempel 1
Started by ●December 10, 2006
Reply by ●December 12, 20062006-12-12
Reply by ●December 12, 20062006-12-12
jim wrote:> > Jerry Avins wrote: > >> You're inferring what OP must have done from your knowledge of what >> can't work. > > Well excuse me, I wasn't aware using knowledge was not permitted. I was > also using the information that his images supplied. After all, if "it > is nothing but a 2D FIR filter", as Rune put it, the images themselves > explain everything. > >> I'm trying to understand what he did from what he said he >> did. I agree that what he said he did can't work, > > No, you are trying to prove your interpretation of what is written on > the software web page can't work.If my interpretation of what he wrote is incorrect, then nothing I wrote on the subject makes any sense. However, his original images aren't out of focus, they're beyond the resolution limit of the camera that made them. That being the case, What does the extra "detail" actually represent? ...> If you are so desperate to know what the impulse response is for the > processing at each of the different parameter settings available why not > download the program and feed it an impulse and see what the response is > for each setting.That's clearly the next step in a rational process. ... Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●December 12, 20062006-12-12
Hi tontoko, Thanks for the explanation. I think a few people are interested in what is written on your web page about the simple 3X3 filters shown there. That is, some people are baffled by the meaning of your comparison of your convolution method to other methods. My interpretation is that no matter how you get there, (what theory and math you choose to use) in the end, the impulse response for a 2d convolution filter that works for correcting the focus on a blurred image is going to look like one strong central positive spike midst a field of smaller negative coefficients. This is inevitable. = I took your web page explanation to mean that other programs simply adjust the relative magnitude of the central spike while your program adjusts the intensity and shape of the surrounding negative coefficients. The part I'm curious about is - does your program do anything to figure out what the ideal impulse response is for a particular blurred image? Or is it up to the skill of the user to come up with the best parameters and then the program simply creates whatever impulse response those parameters define? = -jim tontoko wrote:> => Thank you for your comments and I'm sorry for my response has been > delayed. > => As you've know, my software is an application of the inverse matrix > solution method which is explained in the following site; > => http://139.134.5.123/tiddler2/c22508/iteration.htm > => As the review, when applying Fourier transform to deconvolve the > blurred image, firstly the output image g(x) is formulated with the > true image f(x) as, > => g(x) =3D f*n(x) > => where the extent of n(x) corresponds to the deviation from the focal > point (if g(x) is just in focus, then n(x) =3D ((delta))(x).) After > applying Fourier transform to the above equation, we get; > => G(u) =3D F(u)N(u)> => hence, > => F(u) =3D G(u)/N(u) > => Therefore, by applying inverse Fourier transform to G(u)/N(u), we can > get the image data f(x) before convolution. While my method is to solve=> the first equation: g(x) =3D f*n(x) directly. Firstly that equation is > discretized as, > => g(j) =3D ((sigma))_k f(j-k)n(k) > => It is expressed with matrix-vector form as, > => g =3D Af > => hence, > => f =3D A^(-1)g > => The problem is how to calculate f(x) from the above equation. You may > imagine to apply any standard iterative inverse matrix solution methods=> such as Gauss-Seidel method for it. But we must be careful to apply > Gauss-Seidel method to this specific problem because A doesn't satisfy > the condition which Gauss-Seidel method requires, while my iterative > inverse matrix solution method is enough robust to deal with that > problem (the convergent speed is much slower, though.) > => jim wrote: > > Jerry Avins wrote: > > > > > > > > You're inferring what OP must have done from your knowledge of what=> > > can't work. > > > > Well excuse me, I wasn't aware using knowledge was not permitted. I w=as> > also using the information that his images supplied. After all, if "i=t> > is nothing but a 2D FIR filter", as Rune put it, the images themselve=s> > explain everything. > > > > >I'm trying to understand what he did from what he said he > > > did. I agree that what he said he did can't work, > > > > No, you are trying to prove your interpretation of what is written on=> > the software web page can't work. > > > > > > > and I'l like a clarification from him. > > > > If you are so desperate to know what the impulse response is for the > > processing at each of the different parameter settings available why =not> > download the program and feed it an impulse and see what the response=is> > for each setting. > > > > >Tit you catch Rune's observation that this > > > thread has only half of his remarks? > > > > The OP wrote in a different thread: > > > > > > The pictures shown in the following URL; > > http://www.unmannedspaceflight.com/index.php?act=3DAttach&type=3Dpost=&id=3D2017> > are the original image of Vesta (left), its de-convoluted image (midd=le> > / focus depth:=3D3.8, iteration:=3D8 / software: Focus Corrector ) an=d> > another de-convoluted image (right / focus depth:=3D4.3, iteration:=3D=8 /> > software: Focus Corrector). > > > > > > With his statement "focus depth:=3D4.3, iteration:=3D8" that is enoug=h right> > there to know he is not applying a simple kernel you can write down w=ith> > 3 lines of ascii. So I don't know why you continue to beat that dead > > horse. > > > > -jim > > > > > > > > > Even if you missed the part about there being input paramet=ers and> > > > thought he was using a single filter once only you should know it='s not> > > > that filter that produced those images. Simply by looking at the > > > > pictures you can tell he is using something better than that. Tha=t> > > > filter would leave some blocky artifacts that would be noticeable==2E> > > > Running that filter over and over would simply magnify the blocki=ness.> > > > So your interpretation is clearly wrong. According to Rune that m=akes> > > > you a charlatan because that's what charlatans do - fail to expla=in how> > > > something works. > > > > > > Iterating with that filter is equivalent to a single pass with a la=rger> > > filter, it that's what you mean. > > > > > > > I assume he posted here to get people to download and try t=he program> > > > and offer criticism based on what they find using their own image=s. I> > > > don't think he was looking for critiques on what people would ima=gine> > > > they might find. > > > > > > I'm sure you're right about that. If I get a chance, I'll do that a=nd> > > have a go at the three-image comparison I described. > > > > > > Jerry > > > -- > > > Engineering is the art of making what you want from things you can =get.> > > =AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF==AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF= =AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF> > > > ----=3D Posted via Newsfeeds.Com - Unlimited-Unrestricted-Secure Usen=et News=3D----> > http://www.newsfeeds.com The #1 Newsgroup Service in the World! 120,0=00+ Newsgroups> > --------== Posted via Newsfeeds.Com - Unlimited-Unrestricted-Secure Usenet News==---- http://www.newsfeeds.com The #1 Newsgroup Service in the World! 120,000+ Newsgroups ----= East and West-Coast Server Farms - Total Privacy via Encryption =----
Reply by ●December 12, 20062006-12-12
Jerry Avins wrote:> If my interpretation of what he wrote is incorrect, then nothing I wrot=e> on the subject makes any sense. However, his original images aren't out=> of focus, they're beyond the resolution limit of the camera that made > them. That being the case, What does the extra "detail" actually repres=ent? According to the author's online information: The image out of focus is thought to be deformed from original image by some linear transformation. If you want to write software based on some other assumption(s) go ahead. = -jim> ... > => > If you are so desperate to know what the impulse response is for the > > processing at each of the different parameter settings available why =not> > download the program and feed it an impulse and see what the response=is> > for each setting. > => That's clearly the next step in a rational process. > => ... > => Jerry > -- > Engineering is the art of making what you want from things you can get.=> =AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF==AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF= =AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF ----== Posted via Newsfeeds.Com - Unlimited-Unrestricted-Secure Usenet News==---- http://www.newsfeeds.com The #1 Newsgroup Service in the World! 120,000+ Newsgroups ----= East and West-Coast Server Farms - Total Privacy via Encryption =----
Reply by ●December 12, 20062006-12-12
tontoko wrote:> Thank you for your comments and I'm sorry for my response has been > delayed. > > As you've know, my software is an application of the inverse matrix > solution method which is explained in the following site; > > http://139.134.5.123/tiddler2/c22508/iteration.htm > > As the review, when applying Fourier transform to deconvolve the > blurred image, firstly the output image g(x) is formulated with the > true image f(x) as, > > g(x) = f*n(x) > > where the extent of n(x) corresponds to the deviation from the focal > point (if g(x) is just in focus, then n(x) = ((delta))(x).) After > applying Fourier transform to the above equation, we get; > > G(u) = F(u)N(u) > > hence, > > F(u) = G(u)/N(u) > > Therefore, by applying inverse Fourier transform to G(u)/N(u), we can > get the image data f(x) before convolution. While my method is to solve > the first equation: g(x) = f*n(x) directly. Firstly that equation is > discretized as, > > g(j) = ((sigma))_k f(j-k)n(k) > > It is expressed with matrix-vector form as, > > g = Af > > hence, > > f = A^(-1)g > > The problem is how to calculate f(x) from the above equation. You may > imagine to apply any standard iterative inverse matrix solution methods > such as Gauss-Seidel method for it. But we must be careful to apply > Gauss-Seidel method to this specific problem because A doesn't satisfy > the condition which Gauss-Seidel method requires, while my iterative > inverse matrix solution method is enough robust to deal with that > problem (the convergent speed is much slower, though.)... I completely misunderstood what you were doing. In the words on one comedienne, "Never mind." The images you showed seem to be of diffraction-limited astronomic objects. Do you really penetrate the diffraction-limit barrier? Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●December 13, 20062006-12-13
Jerry Avins wrote: (snip)> I completely misunderstood what you were doing. In the words on one > comedienne, "Never mind." The images you showed seem to be of > diffraction-limited astronomic objects. Do you really penetrate the > diffraction-limit barrier?Deconvolution is an interesting science. I recommend http://www.amazon.com/Deconvolution-Images-Spectra-Peter-Jansson/dp/0123802229 though it was much less than $375 when I bought mine. It has been a while since I read it last, but he has some examples from spectroscopy where a large slit reduces the S/N ratio more than enough to make up for deconvolution on the result. It requires non-linear deconvolution, including the constraint that emission spectra can't go negative, and absorption spectra must be between 0 and 1. -- glen
Reply by ●December 13, 20062006-12-13
I understand that you've asked how to deal with a spike-like signal. Practically when the spike-like signal resides in the center of matrix as, 0 0 0 0 235 0 0 0 0 then the values of pixels around the center are calculated as 235/9, i=2Ee. 26 26 26 26 (235) 26 26 26 26 where the values 235 means the maximum value on RBG spec. As another case, when the spike-like signal resides in the corner of matrix as, 235 0 0 0 0 0 0 0 0 then the values of pixels around the center are calculated as 235-26/9 and -26/9, i.e. 232 -3 -3 -3 (0) -3 -3 -3 -3 If at the end of iterations the value of any pixel is less than zero, its value is substituted with zero. To simplify the explanation, I used here (and in the webpage) the 3x3 matrix, however actually in the software a round disk is used instead of it, which has the radius proportional to the value of the parameter "focus depth". Usual convolution is mathematically interpreted as the conventional iterative inverse matrix solution method, and as mentioned in the webpage, it does not converge to the true image and often causes exaggerated contrast after applying it repeatedly. jim wrote:> Hi tontoko, > Thanks for the explanation. I think a few people are interested in what > is written on your web page about the simple 3X3 filters shown there. > That is, some people are baffled by the meaning of your comparison of > your convolution method to other methods. > > My interpretation is that no matter how you get there, (what theory and > math you choose to use) in the end, the impulse response for a 2d > convolution filter that works for correcting the focus on a blurred > image is going to look like one strong central positive spike midst a > field of smaller negative coefficients. This is inevitable. > I took your web page explanation to mean that other programs simply > adjust the relative magnitude of the central spike while your program > adjusts the intensity and shape of the surrounding negative > coefficients. > > The part I'm curious about is - does your program do anything to figure > out what the ideal impulse response is for a particular blurred image? > Or is it up to the skill of the user to come up with the best parameters > and then the program simply creates whatever impulse response those > parameters define? > > -jim > > tontoko wrote: > > > > Thank you for your comments and I'm sorry for my response has been > > delayed. > > > > As you've know, my software is an application of the inverse matrix > > solution method which is explained in the following site; > > > > http://139.134.5.123/tiddler2/c22508/iteration.htm > > > > As the review, when applying Fourier transform to deconvolve the > > blurred image, firstly the output image g(x) is formulated with the > > true image f(x) as, > > > > g(x) =3D f*n(x) > > > > where the extent of n(x) corresponds to the deviation from the focal > > point (if g(x) is just in focus, then n(x) =3D ((delta))(x).) After > > applying Fourier transform to the above equation, we get; > > > > G(u) =3D F(u)N(u) > > > > > > > > > > > hence, > > > > F(u) =3D G(u)/N(u) > > > > Therefore, by applying inverse Fourier transform to G(u)/N(u), we can > > get the image data f(x) before convolution. While my method is to solve > > the first equation: g(x) =3D f*n(x) directly. Firstly that equation is > > discretized as, > > > > g(j) =3D ((sigma))_k f(j-k)n(k) > > > > It is expressed with matrix-vector form as, > > > > g =3D Af > > > > hence, > > > > f =3D A^(-1)g > > > > The problem is how to calculate f(x) from the above equation. You may > > imagine to apply any standard iterative inverse matrix solution methods > > such as Gauss-Seidel method for it. But we must be careful to apply > > Gauss-Seidel method to this specific problem because A doesn't satisfy > > the condition which Gauss-Seidel method requires, while my iterative > > inverse matrix solution method is enough robust to deal with that > > problem (the convergent speed is much slower, though.) > > > > jim wrote: > > > Jerry Avins wrote: > > > > > > > > > > > You're inferring what OP must have done from your knowledge of what > > > > can't work. > > > > > > Well excuse me, I wasn't aware using knowledge was not permitted. I w=as> > > also using the information that his images supplied. After all, if "it > > > is nothing but a 2D FIR filter", as Rune put it, the images themselves > > > explain everything. > > > > > > >I'm trying to understand what he did from what he said he > > > > did. I agree that what he said he did can't work, > > > > > > No, you are trying to prove your interpretation of what is written on > > > the software web page can't work. > > > > > > > > > > and I'l like a clarification from him. > > > > > > If you are so desperate to know what the impulse response is for the > > > processing at each of the different parameter settings available why =not> > > download the program and feed it an impulse and see what the response=is> > > for each setting. > > > > > > >Tit you catch Rune's observation that this > > > > thread has only half of his remarks? > > > > > > The OP wrote in a different thread: > > > > > > > > > The pictures shown in the following URL; > > > http://www.unmannedspaceflight.com/index.php?act=3DAttach&type=3Dpost=&id=3D2017> > > are the original image of Vesta (left), its de-convoluted image (midd=le> > > / focus depth:=3D3.8, iteration:=3D8 / software: Focus Corrector ) and > > > another de-convoluted image (right / focus depth:=3D4.3, iteration:==3D8 /> > > software: Focus Corrector). > > > > > > > > > With his statement "focus depth:=3D4.3, iteration:=3D8" that is enoug=h right> > > there to know he is not applying a simple kernel you can write down w=ith> > > 3 lines of ascii. So I don't know why you continue to beat that dead > > > horse. > > > > > > -jim > > > > > > > > > > > > Even if you missed the part about there being input paramet=ers and> > > > > thought he was using a single filter once only you should know it='s not> > > > > that filter that produced those images. Simply by looking at the > > > > > pictures you can tell he is using something better than that. That > > > > > filter would leave some blocky artifacts that would be noticeable. > > > > > Running that filter over and over would simply magnify the blocki=ness.> > > > > So your interpretation is clearly wrong. According to Rune that m=akes> > > > > you a charlatan because that's what charlatans do - fail to expla=in how> > > > > something works. > > > > > > > > Iterating with that filter is equivalent to a single pass with a la=rger> > > > filter, it that's what you mean. > > > > > > > > > I assume he posted here to get people to download and try t=he program> > > > > and offer criticism based on what they find using their own image=s=2E I> > > > > don't think he was looking for critiques on what people would ima=gine> > > > > they might find. > > > > > > > > I'm sure you're right about that. If I get a chance, I'll do that a=nd> > > > have a go at the three-image comparison I described. > > > > > > > > Jerry > > > > -- > > > > Engineering is the art of making what you want from things you can =get.> > > > =AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF==AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF= =AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF> > > > > > ----=3D Posted via Newsfeeds.Com - Unlimited-Unrestricted-Secure Usen=et News=3D----> > > http://www.newsfeeds.com The #1 Newsgroup Service in the World! 120,0=00+ Newsgroups> > > ---- > > ----=3D Posted via Newsfeeds.Com - Unlimited-Unrestricted-Secure Usenet N=ews=3D----> http://www.newsfeeds.com The #1 Newsgroup Service in the World! 120,000+ =Newsgroups> ----
