<!doctype html public "-//w3c//dtd html 4.0 transitional//en">
<html>
<p>Jerry Avins wrote:
<blockquote TYPE=CITE>jim wrote:
<br>>
<br>>
<br>> Jerry Avins wrote:
<br>>
<br>>>
<br>>>
<br>>> (Mg+Ye) + (G+Cy) = (G+Cy)+ (Mg+Cy). There's an Mg and a G on each
side.
<br>>> Ye is replaced by an extra Cy on the right.
<br>>>
<br>>>
<br>> Yes that would be a typo it should have been written:
<br>>
<br>> "consecutive samples i.e Y_n = (Mg+Ye) + (G+Cy), Y_n+1= (G+Cy)+
<br>> (Mg+Ye)..."
<br>>
<br>> If we fix that substitute Gr for G in his ascii art it becomes a
little
<br>> more readable and answers most of the questions:
<br>>
<br>> Mg Gr Mg Gr ....
Y Y Y Y....
<br>> Ye Cy Ye Cy...
Y Y Y Y ...
<br>> Gr Mg Gr Mg... =
Y Y Y Y ...
<br>> Ye Cy Ye Cy...
Y Y Y Y ...
<br>> Mg Gr Mg Gr....
Y Y Y Y ...
<p>Then we can make four phantoms where each "real" one would have been
<br>with a one-line delay. (Two-one-line delays if two lines are read out
<br>simultaneously.)
<br> </blockquote>
<p><br>OK, there's that nebulous real again:)
<br> I don't know what his final image resolution is,
but it seems that whether he downsamples from this resolution or not, he
probably needs to filter out the content from fs/2 to fs/4 (not the content
from fs to fs/2 as he specified in his original problem statement). Even
if he doesn't eventually downsample that will smooth out the blockyness
he has now.
<br> The "unique info" (the part of intensity that each
sensor sees alone) in his output array is arranged like this
<p>a1 a1 a2 a2 a3 a3 . . . .
<br>a1 a1 a2 a2 a3 a3 . . . .
<br>b1 b1 b2 b2 b3 b3 . . . .
<br>b1 b1 b2 b2 b3 b3 . . . .
<br>. . . .
<br>
<p>-jim
<br>
<blockquote TYPE=CITE>
<br> ...
<p>Jerry
<br>--
<br>Engineering is the art of making what you want from things you can
get.
<br>¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯</blockquote>
</html>
----== 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 Jerry Avins●January 20, 20062006-01-20
jim wrote:
>
>
> Jerry Avins wrote:
>
>>
>>
>> (Mg+Ye) + (G+Cy) = (G+Cy)+ (Mg+Cy). There's an Mg and a G on each side.
>> Ye is replaced by an extra Cy on the right.
>>
>>
> Yes that would be a typo it should have been written:
>
> "consecutive samples i.e Y_n = (Mg+Ye) + (G+Cy), Y_n+1= (G+Cy)+
> (Mg+Ye)..."
>
> If we fix that substitute Gr for G in his ascii art it becomes a little
> more readable and answers most of the questions:
>
> Mg Gr Mg Gr .... Y Y Y Y....
> Ye Cy Ye Cy... Y Y Y Y ...
> Gr Mg Gr Mg... = Y Y Y Y ...
> Ye Cy Ye Cy... Y Y Y Y ...
> Mg Gr Mg Gr.... Y Y Y Y ...
Then we can make four phantoms where each "real" one would have been
with a one-line delay. (Two-one-line delays if two lines are read out
simultaneously.)
...
Jerry
--
Engineering is the art of making what you want from things you can get.
�����������������������������������������������������������������������
Reply by jim●January 20, 20062006-01-20
<!doctype html public "-//w3c//dtd html 4.0 transitional//en">
<html>
<p>Jerry Avins wrote:
<blockquote TYPE=CITE>
<p>(Mg+Ye) + (G+Cy) = (G+Cy)+ (Mg+Cy). There's an Mg and a G on each side.
<br>Ye is replaced by an extra Cy on the right.
<br> </blockquote>
Yes that would be a typo it should have been written:
<p>"consecutive samples i.e Y_n = (Mg+Ye) + (G+Cy), Y_n+1= (G+Cy)+
(Mg+Ye)..."
<p>If we fix that substitute Gr for G in his ascii art it becomes a little
more readable and answers most of the questions:
<p>Mg Gr Mg Gr ....
Y Y Y Y....
<br>Ye Cy Ye Cy...
Y Y Y Y ...
<br>Gr Mg Gr Mg.... = Y Y Y Y
...
<br>Ye Cy Ye Cy...
Y Y Y Y ...
<br>Mg Gr Mg Gr....
Y Y Y Y ...
<p>-jim
<br>
<blockquote TYPE=CITE>
<br>> Each scan line is not the same. What I wasn't
<br>> able to determine was is the output array the same size as all the
<br>> sensors combined? or is it 1/4 the size of the sensor total? I was
<br>> guessing the latter. And also there is the question, are there equal
<br>> numbers of each type of sensor? or are there different effective
sample
<br>> rates for each color type involved? It's also not clear whether he
is
<br>> computing intensity for purposes of producing a grayscale image or
if
<br>> it's just part of the process of converting to a different color
space
<br>> like HSB.
<br>>
<br>>
<br>>>
<br>>>
<br>>> I don't know enough about the structure of his sensor to suggest
a
<br>>> scheme, but I think that creating phantom pixels is the right first
<br>>> step. For a stripe pattern of ... R1 G1 B1 R2 G2 B2 ... and assuming
<br>>> that the gains (weights) have already been adjusted, I would create
<br>>> grayscale pixels ... G1 B1 R2, B1 R2 G2, R2 G2 B2, ... so that each
<br>>> color stripe contributes to three pixels, then low-pass to 1/3 the
<br>>> phantom pixel rate. With details, we could probably work out a similar
<br>>> "oversampling" scheme for his line sensor.
<br>>>
<br>>>
<br>>
<br>> I don't know if thinking in terms of RGB is helpful here since that
<br>> isn't the colorspace he has.
<p>I wanted only to illustrate what I meant by phantom pixels. With his
<br>sensor, he might need a one-scan-line delay to make them. I picked
a
<br>simple case for illustration.
<p>>
Anyway, what is at issue is sample rates.
<p>Yes. we agree on that and most of everything else. Maybe guarev will
<br>offer specs for the sensor array.
<p> ...
<p>Jerry
<br>--
<br>Engineering is the art of making what you want from things you can
get.
<br>¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯</blockquote>
</html>
----== 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 Jerry Avins●January 20, 20062006-01-20
jim wrote:
>
>
> Jerry Avins wrote:
...
>> them. (Is his "Y = (Mg+Ye) + (G+Cy) = (G+Cy)+ (Mg+Cy)..." right?)
>>
>
> Yes, probably correct.
(Mg+Ye) + (G+Cy) = (G+Cy)+ (Mg+Cy). There's an Mg and a G on each side.
Ye is replaced by an extra Cy on the right.
> Each scan line is not the same. What I wasn't
> able to determine was is the output array the same size as all the
> sensors combined? or is it 1/4 the size of the sensor total? I was
> guessing the latter. And also there is the question, are there equal
> numbers of each type of sensor? or are there different effective sample
> rates for each color type involved? It's also not clear whether he is
> computing intensity for purposes of producing a grayscale image or if
> it's just part of the process of converting to a different color space
> like HSB.
>
>
>>
>>
>> I don't know enough about the structure of his sensor to suggest a
>> scheme, but I think that creating phantom pixels is the right first
>> step. For a stripe pattern of ... R1 G1 B1 R2 G2 B2 ... and assuming
>> that the gains (weights) have already been adjusted, I would create
>> grayscale pixels ... G1 B1 R2, B1 R2 G2, R2 G2 B2, ... so that each
>> color stripe contributes to three pixels, then low-pass to 1/3 the
>> phantom pixel rate. With details, we could probably work out a similar
>> "oversampling" scheme for his line sensor.
>>
>>
>
> I don't know if thinking in terms of RGB is helpful here since that
> isn't the colorspace he has.
I wanted only to illustrate what I meant by phantom pixels. With his
sensor, he might need a one-scan-line delay to make them. I picked a
simple case for illustration.
> Anyway, what is at issue is sample rates.
Yes. we agree on that and most of everything else. Maybe guarev will
offer specs for the sensor array.
...
Jerry
--
Engineering is the art of making what you want from things you can get.
�����������������������������������������������������������������������
Reply by jim●January 20, 20062006-01-20
<!doctype html public "-//w3c//dtd html 4.0 transitional//en">
<html>
<p>Jerry Avins wrote:
<blockquote TYPE=CITE>jim wrote:
<p> ...
<p>> I don't think you're really grasping the problem. What he has is a
<br>> sampling array that has 4 times as many sample points as the output
<br>> array. So the problem would seem to be a simple matter of devising
a
<br>> filter suitable for downsampling by 4. The kicker is that each of
these
<br>> four receptors aren't really measuring the same thing.
<br>> The conventional way of looking at it would
be that each of the 4
<br>> receptors only 'sees' 1/4 of the signal (or some other division summing
<br>> to unity) so you simply sum the outputs (or do weighted sum). Its
like
<br>> having 4 judges with different kind of color blindness voting to
tell
<br>> you what the grayscale brightness is. But there is probably more
to it
<br>> than that. For one thing his way of looking at it takes into account
<br>> that the 4 receptors are not in the same location. Plus there is
some
<br>> overlap in the light frequencies that each receptor is sensitive
to
<br>> (note: that is a different sort of frequency than spatial frequency).
So
<br>> some sort of filtering could be better than simply naively adding
the
<br>> output of blocks of 4.
<p>Another aspect of the problem is that the sensors being smaller than
the
<br>cell size, the inherent filtering of blocky arrays is lost. There is
a
<br>way to oversample the array that will make the filtering he wants
<br>possible, but I can't say it will help. First of all, when he says
<br>"halfband filter" he seems to mean the Nyquist rate for the pixel size.
<br>Before he can filter out those higher frequencies, he has to create
<br>them. (Is his "Y = (Mg+Ye) + (G+Cy) = (G+Cy)+ (Mg+Cy)..." right?)</blockquote>
<p><br>Yes, probably correct. Each scan line is not the same. What I wasn't
able to determine was is the output array the same size as all the sensors
combined? or is it 1/4 the size of the sensor total? I was guessing the
latter. And also there is the question, are there equal numbers of each
type of sensor? or are there different effective sample rates for each
color type involved? It's also not clear whether he is computing intensity
for purposes of producing a grayscale image or if it's just part of the
process of converting to a different color space like HSB.
<br>
<blockquote TYPE=CITE>
<p>I don't know enough about the structure of his sensor to suggest a
<br>scheme, but I think that creating phantom pixels is the right first
<br>step. For a stripe pattern of ... R1 G1 B1 R2 G2 B2 ... and assuming
<br>that the gains (weights) have already been adjusted, I would create
<br>grayscale pixels ... G1 B1 R2, B1 R2 G2, R2 G2 B2, ... so that each
<br>color stripe contributes to three pixels, then low-pass to 1/3 the
<br>phantom pixel rate. With details, we could probably work out a similar
<br>"oversampling" scheme for his line sensor.
<br> </blockquote>
<p><br>I don't know if thinking in terms of RGB is helpful here since that
isn't the colorspace he has. Anyway, what is at issue is sample rates.
Think of it this way: there are 2 types of information that each sensor
receives. There is unique info and shared info. Unique info is lightwaves
that only one channel receives. Shared info is lightwaves that 2 or more
channels 'sees'. Now its easy to see that shared info is sampled at a higher
effective rate than unique info. If no shared info exists then you might
as well just sum the channels as any damage done by aliasing can't be helped.
So any filtering scheme would seem to be targeted at preventing aliasing
of the shared info which has a higher sample rate.
<br>
<blockquote TYPE=CITE>
<br>This whole discussion hinges on what the sample rate really is. Maybe
<br>when it's all matrixed out, it really does boil down to an FIR running
<br>at the 4x rate, provided the signals are weighted beforehand. I don't
<br>know. I do know that FIRs can be flat enough and can have all the
<br>sharpness needed if latency isn't a problem. I don't see one scan line
<br>of latency being a problem for most uses, and that many taps would
make
<br>a jim dandy filter if coefficient truncation doesn't mess it up too
badly.</blockquote>
<p><br>I don't think plugging in flatness and transition sharpness into
a filter program is going to help much here. For one thing humans can only
perceive about 6-bits worth of brightness levels so long filters are going
to create a lot more problems than they solve. Human visual response isn't
flat either so using that as a metric of filter design doesn't make much
sense either. Also it probably is useful to view this as a multi-rate filtering
problem. That is, each channel has its own set of coefficients that contribute
to each output pixel. There may even be multi-rates within channels. It
looks like his pattern may repeat every other line so there might be for
each channel 2 sets of coefficients depending on which line is being processed.
<p>-jim
<br>
<br>
<br>
<blockquote TYPE=CITE>
<p>Jerry
<p>Engineering is the art of making what you want from things you can get.
<br>¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯</blockquote>
</html>
----== 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 Jerry Avins●January 20, 20062006-01-20
jim wrote:
...
> I don't think you're really grasping the problem. What he has is a
> sampling array that has 4 times as many sample points as the output
> array. So the problem would seem to be a simple matter of devising a
> filter suitable for downsampling by 4. The kicker is that each of these
> four receptors aren't really measuring the same thing.
> The conventional way of looking at it would be that each of the 4
> receptors only 'sees' 1/4 of the signal (or some other division summing
> to unity) so you simply sum the outputs (or do weighted sum). Its like
> having 4 judges with different kind of color blindness voting to tell
> you what the grayscale brightness is. But there is probably more to it
> than that. For one thing his way of looking at it takes into account
> that the 4 receptors are not in the same location. Plus there is some
> overlap in the light frequencies that each receptor is sensitive to
> (note: that is a different sort of frequency than spatial frequency). So
> some sort of filtering could be better than simply naively adding the
> output of blocks of 4.
Another aspect of the problem is that the sensors being smaller than the
cell size, the inherent filtering of blocky arrays is lost. There is a
way to oversample the array that will make the filtering he wants
possible, but I can't say it will help. First of all, when he says
"halfband filter" he seems to mean the Nyquist rate for the pixel size.
Before he can filter out those higher frequencies, he has to create
them. (Is his "Y = (Mg+Ye) + (G+Cy) = (G+Cy)+ (Mg+Cy)..." right?)
I don't know enough about the structure of his sensor to suggest a
scheme, but I think that creating phantom pixels is the right first
step. For a stripe pattern of ... R1 G1 B1 R2 G2 B2 ... and assuming
that the gains (weights) have already been adjusted, I would create
grayscale pixels ... G1 B1 R2, B1 R2 G2, R2 G2 B2, ... so that each
color stripe contributes to three pixels, then low-pass to 1/3 the
phantom pixel rate. With details, we could probably work out a similar
"oversampling" scheme for his line sensor.
This whole discussion hinges on what the sample rate really is. Maybe
when it's all matrixed out, it really does boil down to an FIR running
at the 4x rate, provided the signals are weighted beforehand. I don't
know. I do know that FIRs can be flat enough and can have all the
sharpness needed if latency isn't a problem. I don't see one scan line
of latency being a problem for most uses, and that many taps would make
a jim dandy filter if coefficient truncation doesn't mess it up too badly.
Jerry
Engineering is the art of making what you want from things you can get.
�����������������������������������������������������������������������
Reply by jim●January 19, 20062006-01-19
<!doctype html public "-//w3c//dtd html 4.0 transitional//en">
<html>
<p>Jerry Avins wrote:
<blockquote TYPE=CITE>gaurav811 wrote:
<br>> Thank you all for your response. I will elaborate now further what
I want
<br>> to do. I hope this time its all complete.
<br>> Now I have a Sony CCD that has CMYG color filters placed in following
<br>> Mosaic pattern.
<br>>
<br>> Mg G Mg G .... Y
Y Y Y....
<br>> Ye Cy Ye Cy... Y
Y Y Y ...
<br>> G Mg G Mg.... = Y Y Y Y
...
<br>> Ye Cy Ye Cy... Y
Y Y Y ...
<br>> Mg G Mg G....
Y Y Y Y ...
<br>> . . .
<br>> . . .
<br>> .and so on
<br>>
<br>> Now the first line of the image generated from the above mosaic pattern
<br>> will be addition of the two horizontal row values i.e (Mg+Ye), (G+Cy),
<br>> (Mg+Ye), (G+Cy) ... and so on
<br>> and second line would be (G+Ye), (Mg+Cy), (G+Ye) ... and so on.
<br>>
<br>> This signal would be analog and would be first converted to digital
form
<br>> by an A2D. To generate the luminance signal I need to "ADD" the
<br>> consecutive samples i.e Y = (Mg+Ye) + (G+Cy) = (G+Cy)+ (Mg+Cy)...
<br>>
<br>> The reason for putting ADD in quotation is because this addition
would be
<br>> like a low pass filtering by an FIR with two equal coeffs having
<br>> appropriate scaling. Now this filter response actually has a very
wide
<br>> transition band and hence will bring a lot of high frequency component
i.e
<br>> has a very low stop band attenuation and high transition width.
<p>If it's already digital, there are no high frequency components to be
<br>filtered. They're all aliased below Fs/2.</blockquote>
<p><br>I don't think you're really grasping the problem. What he has is
a sampling array that has 4 times as many sample points as the output array.
So the problem would seem to be a simple matter of devising a filter suitable
for downsampling by 4. The kicker is that each of these four receptors
aren't really measuring the same thing.
<br> The conventional way of looking at it would be that
each of the 4 receptors only 'sees' 1/4 of the signal (or some other division
summing to unity) so you simply sum the outputs (or do weighted sum). Its
like having 4 judges with different kind of color blindness voting to tell
you what the grayscale brightness is. But there is probably more to it
than that. For one thing his way of looking at it takes into account that
the 4 receptors are not in the same location. Plus there is some overlap
in the light frequencies that each receptor is sensitive to (note: that
is a different sort of frequency than spatial frequency). So some sort
of filtering could be better than simply naively adding the output of blocks
of 4.
<p>-jim
<br>
<br>
<br>
<br>
<br>
<blockquote TYPE=CITE>
<p>> Thus after the addition process I want to filter this signal with
a half
<br>> band low pass filter to remove the unwanted high frequency component.
<p>A half-band filter is symmetric about Fs/4. There are no components
<br>above half the sample rate that are distinct from those below.
<p>> I am
<br>> currently not very sure about why we need a sharp FIR filter but
I am told
<br>> that since the Cyan Magenta Green and Yellow have their own seperate
<br>> wavelengths and so the process of adding them leads to spurious intensity
<br>> values at the high frequencies. Well so this is the purpose of the
<br>> excercise. I am sure nobody would be absolutely clear with this
<br>> explanation in regards to why I need such a filter but you should
know now
<br>> what I am trying to do here.
<br>>
<br>> For now I am going to try the standard FIR`s available in Matlab
and if
<br>> boss doesent like the Filter response will try to use the Frequency
<br>> Masking Technique. But in that technique I really do not know how
the
<br>> bandedge filter gets implemented in real time. I am refering to "
A new
<br>> approach for design sharp filters using frequency response masking
<br>> technique" by Lei Zhang and Yong Lian. I dont know if we replace
the
<br>> normal delays in the bandedge filters by M delays what happens to
the data
<br>> coming at the input rate. I mean if we put M delays at each tap then
the
<br>> total delay in the filter would be M*size of filter. So to implement
this
<br>> in real time either the filter should run at M times the clock rate
or
<br>> there can be some other solution which I dont know off.
<br>>
<br>> Thanks a lot for your help.I will update again as soon as I find
more
<br>> about this problem,
<br>> Gaurav
<p>You are trying to filter a signal that isn't there. Like in
<br>Winnie-the-Pooh "Where the Woozle Wasn't", that's going around in circles.
<p>Jerry
<br>--
<br>Engineering is the art of making what you want from things you can
get.
<br>¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯</blockquote>
</html>
----== 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 Jerry Avins●January 19, 20062006-01-19
gaurav811 wrote:
> Thank you all for your response. I will elaborate now further what I want
> to do. I hope this time its all complete.
> Now I have a Sony CCD that has CMYG color filters placed in following
> Mosaic pattern.
>
> Mg G Mg G .... Y Y Y Y....
> Ye Cy Ye Cy... Y Y Y Y ...
> G Mg G Mg.... = Y Y Y Y ...
> Ye Cy Ye Cy... Y Y Y Y ...
> Mg G Mg G.... Y Y Y Y ...
> . . .
> . . .
> .and so on
>
> Now the first line of the image generated from the above mosaic pattern
> will be addition of the two horizontal row values i.e (Mg+Ye), (G+Cy),
> (Mg+Ye), (G+Cy) ... and so on
> and second line would be (G+Ye), (Mg+Cy), (G+Ye) ... and so on.
>
> This signal would be analog and would be first converted to digital form
> by an A2D. To generate the luminance signal I need to "ADD" the
> consecutive samples i.e Y = (Mg+Ye) + (G+Cy) = (G+Cy)+ (Mg+Cy)...
>
> The reason for putting ADD in quotation is because this addition would be
> like a low pass filtering by an FIR with two equal coeffs having
> appropriate scaling. Now this filter response actually has a very wide
> transition band and hence will bring a lot of high frequency component i.e
> has a very low stop band attenuation and high transition width.
If it's already digital, there are no high frequency components to be
filtered. They're all aliased below Fs/2.
> Thus after the addition process I want to filter this signal with a half
> band low pass filter to remove the unwanted high frequency component.
A half-band filter is symmetric about Fs/4. There are no components
above half the sample rate that are distinct from those below.
> I am
> currently not very sure about why we need a sharp FIR filter but I am told
> that since the Cyan Magenta Green and Yellow have their own seperate
> wavelengths and so the process of adding them leads to spurious intensity
> values at the high frequencies. Well so this is the purpose of the
> excercise. I am sure nobody would be absolutely clear with this
> explanation in regards to why I need such a filter but you should know now
> what I am trying to do here.
>
> For now I am going to try the standard FIR`s available in Matlab and if
> boss doesent like the Filter response will try to use the Frequency
> Masking Technique. But in that technique I really do not know how the
> bandedge filter gets implemented in real time. I am refering to " A new
> approach for design sharp filters using frequency response masking
> technique" by Lei Zhang and Yong Lian. I dont know if we replace the
> normal delays in the bandedge filters by M delays what happens to the data
> coming at the input rate. I mean if we put M delays at each tap then the
> total delay in the filter would be M*size of filter. So to implement this
> in real time either the filter should run at M times the clock rate or
> there can be some other solution which I dont know off.
>
> Thanks a lot for your help.I will update again as soon as I find more
> about this problem,
> Gaurav
You are trying to filter a signal that isn't there. Like in
Winnie-the-Pooh "Where the Woozle Wasn't", that's going around in circles.
Jerry
--
Engineering is the art of making what you want from things you can get.
�����������������������������������������������������������������������
Reply by jim●January 19, 20062006-01-19
<!doctype html public "-//w3c//dtd html 4.0 transitional//en">
<html>
<p>gaurav811 wrote:
<blockquote TYPE=CITE>Thank you all for your response. I will elaborate
now further what I want
<br>to do. I hope this time its all complete.
<br>Now I have a Sony CCD that has CMYG color filters placed in following
<br>Mosaic pattern.
<p>Mg G Mg G .... Y Y Y
Y....
<br>Ye Cy Ye Cy... Y Y
Y Y ...
<br>G Mg G Mg.... = Y Y Y Y ...
<br>Ye Cy Ye Cy... Y Y
Y Y ...
<br>Mg G Mg G....
Y Y Y Y ...
<br> . . .
<br> . . .
<br>.and so on
<p>Now the first line of the image generated from the above mosaic pattern
<br>will be addition of the two horizontal row values i.e (Mg+Ye), (G+Cy),
<br>(Mg+Ye), (G+Cy) ... and so on
<br>and second line would be (G+Ye), (Mg+Cy), (G+Ye) ... and so on.
<p>This signal would be analog and would be first converted to digital
form
<br>by an A2D. To generate the luminance signal I need to "ADD" the
<br>consecutive samples i.e Y = (Mg+Ye) + (G+Cy) = (G+Cy)+ (Mg+Cy)...</blockquote>
So what you are doing is converting a color scheme to a grayscale scheme.
I don't know that any one has viewed that as a convolution operation before.
If you were to just store each color separately each in its own 2d array
would the 4 arrays be equal in size or is the distribution of the different
receptors different?
<br>
<blockquote TYPE=CITE>
<p>The reason for putting ADD in quotation is because this addition would
be
<br>like a low pass filtering by an FIR with two equal coeffs having
<br>appropriate scaling. Now this filter response actually has a very wide
<br>transition band and hence will bring a lot of high frequency component
i.e
<br>has a very low stop band attenuation and high transition width.
<br> </blockquote>
Well you could look at it that way if they were all measuring the same
thing. But they aren't so I don't think you can view 2 colors next to each
other which have very different values as being indicative of high frequency.
But it's an interesting idea and may have some validity since you are only
interested in the grayscale output. But an image with only one hue
that is made up of 100% of 2 of the colors and zero of the other 2 colors
would be a very high frequency image by your reasoning. But of course addition
is a good enough filter to remove that high frequency completely. I don't
know, seems like addition has exactly the right frequency response for
what you want.
<br>
<br>
<blockquote TYPE=CITE>
<br>Thus after the addition process I want to filter this signal with a
half
<br>band low pass filter to remove the unwanted high frequency component.
I am
<br>currently not very sure about why we need a sharp FIR filter but I
am told
<br>that since the Cyan Magenta Green and Yellow have their own seperate
<br>wavelengths and so the process of adding them leads to spurious intensity
<br>values at the high frequencies. Well so this is the purpose of the
<br>excercise. I am sure nobody would be absolutely clear with this
<br>explanation in regards to why I need such a filter but you should know
now
<br>what I am trying to do here.</blockquote>
<p><br>You may want to remove high frequencies. But I'm not sure to what
degree mixing the different channels has contributed to producing them.
Usually when converting from color to grayscale you weight the different
color channels according to how the human eyeball perceives the intensity
from each color (or at least what people believe that to be). But if there
is any validity to what you are saying you don't want to do the addition
and then filter, you want to filter instead of the addition as if you were
down sampling by 4.
<p>-jim
<br>
<blockquote TYPE=CITE>
<p>For now I am going to try the standard FIR`s available in Matlab and
if
<br>boss doesent like the Filter response will try to use the Frequency
<br>Masking Technique. But in that technique I really do not know how the
<br>bandedge filter gets implemented in real time. I am refering to " A
new
<br>approach for design sharp filters using frequency response masking
<br>technique" by Lei Zhang and Yong Lian. I dont know if we replace the
<br>normal delays in the bandedge filters by M delays what happens to the
data
<br>coming at the input rate. I mean if we put M delays at each tap then
the
<br>total delay in the filter would be M*size of filter. So to implement
this
<br>in real time either the filter should run at M times the clock rate
or
<br>there can be some other solution which I dont know off.
<p>Thanks a lot for your help.I will update again as soon as I find more
<br>about this problem,
<br>Gaurav
<p>>jim wrote:
<br>>
<br>> ...
<br>>> ... If you are trying to build an anti-alias
filter -
<br>>> you're too late you can't anti-alias filter after it's digital.
<br>>
<br>>How dumb of me! I ought to have realized that that's what gaurav wanted
<br>>to do.
<br>>
<br>>gaurav:
<br>>
<br>>Any frequencies above Nyquist will be aliased into the band below
by the
<p>>sampling process. That's why they are removed before sampling when
<br>>that's possible. After sampling, there are no components that can
be
<br>>said to be above Fs/2; they've already been aliased below.
<br>>
<br>>Digital images are filtered before sampling by two processes, neither
<br>>very good, but they're all we have. First, high-frequencies are
<br>>attenuated by the averaging effect of finite-size pixel sensors. Second,
<p>>they are attenuated by the ability of the imaging system; sharper (more
<br>>highs) is not always better. I have seen diffusers used, and deliberate
<br>>defocussing. I'm not in a position to say that either way to filter
is a
<p>>good idea.
<br>>
<br>>Jerry
<br>>--
<br>>Engineering is the art of making what you want from things you can
get.
<br>>¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
<br>></blockquote>
</html>
----== 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 gaurav811●January 19, 20062006-01-19
One mistake in the explanation. Actually I have to design a filter to
replace the addition process. Basically instead of just adding consecutive
values I have to design a half band filter and use that to filter the
signal so that it has a better frequency response compared to the
addition.
Off Topic: This might be of interest to you. Its something I did for a
project.
http://www.ic.sunysb.edu/stu/gsharma/Filter%20Design%20Report.pdf
>Thank you all for your response. I will elaborate now further what I
want
>to do. I hope this time its all complete.
>Now I have a Sony CCD that has CMYG color filters placed in following
>Mosaic pattern.
>
>Mg G Mg G .... Y Y Y Y....
>Ye Cy Ye Cy... Y Y Y Y ...
>G Mg G Mg.... = Y Y Y Y ...
>Ye Cy Ye Cy... Y Y Y Y ...
>Mg G Mg G.... Y Y Y Y ...
> . . .
> . . .
>.and so on
>
>Now the first line of the image generated from the above mosaic pattern
>will be addition of the two horizontal row values i.e (Mg+Ye), (G+Cy),
>(Mg+Ye), (G+Cy) ... and so on
>and second line would be (G+Ye), (Mg+Cy), (G+Ye) ... and so on.
>
>This signal would be analog and would be first converted to digital form
>by an A2D. To generate the luminance signal I need to "ADD" the
>consecutive samples i.e Y = (Mg+Ye) + (G+Cy) = (G+Cy)+ (Mg+Cy)...
>
>The reason for putting ADD in quotation is because this addition would
be
>like a low pass filtering by an FIR with two equal coeffs having
>appropriate scaling. Now this filter response actually has a very wide
>transition band and hence will bring a lot of high frequency component
i.e
>has a very low stop band attenuation and high transition width.
>
>Thus after the addition process I want to filter this signal with a half
>band low pass filter to remove the unwanted high frequency component. I
am
>currently not very sure about why we need a sharp FIR filter but I am
told
>that since the Cyan Magenta Green and Yellow have their own seperate
>wavelengths and so the process of adding them leads to spurious
intensity
>values at the high frequencies. Well so this is the purpose of the
>excercise. I am sure nobody would be absolutely clear with this
>explanation in regards to why I need such a filter but you should know
now
>what I am trying to do here.
>
>For now I am going to try the standard FIR`s available in Matlab and if
>boss doesent like the Filter response will try to use the Frequency
>Masking Technique. But in that technique I really do not know how the
>bandedge filter gets implemented in real time. I am refering to " A new
>approach for design sharp filters using frequency response masking
>technique" by Lei Zhang and Yong Lian. I dont know if we replace the
>normal delays in the bandedge filters by M delays what happens to the
data
>coming at the input rate. I mean if we put M delays at each tap then the
>total delay in the filter would be M*size of filter. So to implement
this
>in real time either the filter should run at M times the clock rate or
>there can be some other solution which I dont know off.
>
>Thanks a lot for your help.I will update again as soon as I find more
>about this problem,
>Gaurav
>
>
>>jim wrote:
>>
>> ...
>>> ... If you are trying to build an anti-alias filter -
>>> you're too late you can't anti-alias filter after it's digital.
>>
>>How dumb of me! I ought to have realized that that's what gaurav wanted
>>to do.
>>
>>gaurav:
>>
>>Any frequencies above Nyquist will be aliased into the band below by
the
>
>>sampling process. That's why they are removed before sampling when
>>that's possible. After sampling, there are no components that can be
>>said to be above Fs/2; they've already been aliased below.
>>
>>Digital images are filtered before sampling by two processes, neither
>>very good, but they're all we have. First, high-frequencies are
>>attenuated by the averaging effect of finite-size pixel sensors.
Second,
>
>>they are attenuated by the ability of the imaging system; sharper (more
>>highs) is not always better. I have seen diffusers used, and deliberate
>>defocussing. I'm not in a position to say that either way to filter is
a
>
>>good idea.
>>
>>Jerry
>>--
>>Engineering is the art of making what you want from things you can get.
>>�����������������������������������������������������������������������
>>
>
>
>