imageop suggestion

[jo...@verona.se]

The imageop block is quite useful for image operations, much more so
than the imagereader in the core. It would be nice if it could be
released.

However there seems to be some problems with rotation code.
I mostly managed to get lots of black areas, and spurious image
fragments in the rotated image.

I had a look at the cocde and its not obvious what a rotation should
do, which I think is the root cause of the problem. (that is, what
should be the bounding box, where should the anchor for rotation be etc)

I propose hiding the general rotation operator, and implementing a
rotate-at-right-angle operator.

This special case is IMHO the most useful type of rotation, its
obvious how it should work, and could be more easily implemented and
perhaps optimized.

I can implement this if the list decides this is a good idea. (I will
probably anyway for my own needs). The general rotation operator could
be tackledd later, but wouldnt be necessary for release.


-- 
Joakim Verona

Re: imageop suggestion

[jo...@verona.se]

Grzegorz Kossakowski <gk...@apache.org> writes:

> joakim@verona.se pisze:
>> The imageop block is quite useful for image operations, much more so
>> than the imagereader in the core. It would be nice if it could be
>> released.
>>
>> However there seems to be some problems with rotation code.
>> I mostly managed to get lots of black areas, and spurious image
>> fragments in the rotated image.
>>
>> I had a look at the cocde and its not obvious what a rotation should
>> do, which I think is the root cause of the problem. (that is, what
>> should be the bounding box, where should the anchor for rotation be etc)
>>
>> I propose hiding the general rotation operator, and implementing a
>> rotate-at-right-angle operator.
>
> Can you explain how this operator would work? I don't have much experience with image processing so I may miss something obvious.

I tried to find a name that would be obvious, I dont know if I
suceeded. Another name might be "order-4-rotate".

right-angle-rotate simply means rotate by 90, 180, or 270 degrees, and
dont accept intermediary values.

The current rotate operator is supposed to be able to do this, but
somehow fails. 


>
>> This special case is IMHO the most useful type of rotation, its
>> obvious how it should work, and could be more easily implemented and
>> perhaps optimized.
>>
>> I can implement this if the list decides this is a good idea. (I will
>> probably anyway for my own needs). The general rotation operator could
>> be tackledd later, but wouldnt be necessary for release.
>
> Feel free to submit patch that will include this new operator,
> adjusted sample and test case. I'm very happy that you want to spend
> some time on imageop block and make it ready for a release.

I will try next week, but I will be on a 3g modem, so image
trial-and-terror will be somewhat slow. I already have something
somewhat working, but need to test all the cases.



>
> -- 
> Grzegorz Kossakowski
> http://reflectingonthevicissitudes.wordpress.com/

-- 
Joakim Verona

Re: imageop suggestion

[Grzegorz Kossakowski <gk...@apache.org>]

joakim@verona.se pisze:
> The imageop block is quite useful for image operations, much more so
> than the imagereader in the core. It would be nice if it could be
> released.
> 
> However there seems to be some problems with rotation code.
> I mostly managed to get lots of black areas, and spurious image
> fragments in the rotated image.
> 
> I had a look at the cocde and its not obvious what a rotation should
> do, which I think is the root cause of the problem. (that is, what
> should be the bounding box, where should the anchor for rotation be etc)
> 
> I propose hiding the general rotation operator, and implementing a
> rotate-at-right-angle operator.

Can you explain how this operator would work? I don't have much experience with image processing so I may miss something obvious.

> This special case is IMHO the most useful type of rotation, its
> obvious how it should work, and could be more easily implemented and
> perhaps optimized.
> 
> I can implement this if the list decides this is a good idea. (I will
> probably anyway for my own needs). The general rotation operator could
> be tackledd later, but wouldnt be necessary for release.

Feel free to submit patch that will include this new operator, adjusted sample and test case. I'm very happy that you want to spend some 
time on imageop block and make it ready for a release.

-- 
Grzegorz Kossakowski
http://reflectingonthevicissitudes.wordpress.com/