Squeak and Tesseract
Locked
 
|
Hello list, As I'm looking at a way to automate the search of documents in my humble administration, I read some articles about OCR. I came along an article about using Python with Tesseract, to transform an scan of a document into text, that is searchable. My question now is if I can do something similar with Squeak. To my inexperienced eye, it seems like I should use FFI to call the functions in the Tesseract API, but this API is in C++, and I don't know if it is possible to use FFI to call C++ functions? Or should I forget the API and use OSProcess to start the tesseract program? Could anyone point me in the right direction, or just tell if the whole idea is insane? But if it can be done in Python.... Thanks in advance, Edwin Ancaer. |
|
|
On Fri, 2 Nov 2018 at 18:44, Edwin Ancaer <[hidden email]> wrote:
You are right C++ is difficult because of the name mangling of function symbols, but good fortune I notice Tesseract has C bindings... so it looks like you are in the clear.
FFI will be more flexible.
I think its a great idea and actually Tesseract FFI is something I've wanted to play with before but not had the time. I'd be interested to hear how you go with it. cheers -ben |
|
Administrator
|
In reply to this post by Edwin Ancaer
Edwin Ancaer wrote
> it seems like I should use FFI to call the functions in > the Tesseract API, but this API is in C++, and I don't know if it is > possible to use FFI to call C++ functions? The typical workaround is to wrap the API you want to use in a C library that exposes them unmangled. Edwin Ancaer wrote > Or should I forget the API and use OSProcess to start the tesseract > program? I took this approach for Pharo: https://github.com/seandenigris/Tesseract-St . There likely will be a lot there that you can use in Squeak. ----- Cheers, Sean -- Sent from: http://forum.world.st/Squeak-Dev-f45488.html
Cheers,
Sean |
|
|
> On 2018-11-02, at 7:28 PM, Sean P. DeNigris <[hidden email]> wrote: > > Edwin Ancaer wrote >> it seems like I should use FFI to call the functions in >> the Tesseract API, but this API is in C++, and I don't know if it is >> possible to use FFI to call C++ functions? > > The typical workaround is to wrap the API you want to use in a C library > that exposes them unmangled. Also consider the trick used in the Bochs plugin - extern "C" { code....} See https://stackoverflow.com/questions/1041866/what-is-the-effect-of-extern-c-in-c for a faintly intelligible discussion tim -- tim Rowledge; [hidden email]; http://www.rowledge.org/tim "!" The strange little noise you make when you can't scream... |
|
|
In reply to this post by Ben Coman
While I've done a lot of C programming that is useful for FFI interfacing, I've not done much C++. So just sharing something new I learnt today to help with FFI interfacing to combined C/C++ libraries. I thought maybe others in the same boat could be interested in this. [Original question asked in squeak-dev, cross-posting to pharo-dev] On Fri, 2 Nov 2018 at 21:06, Ben Coman <[hidden email]> wrote:
Browsing a deeper I got quite confused for a while. I could see a typedef definition for TessResultRenderer here... https://github.com/tesseract-ocr/tesseract/blob/master/src/api/capi.h#L83 "typedef struct TessResultRenderer TessResultRenderer" which I understood to must refer to *existing* struct, but I couldn't find the definition of that struct anywhere. In particular... $ git clone [hidden email]:tesseract-ocr/tesseract.git $ cd tesseract $ find . -type f -name "*h" -exec grep -Hn TessResultRenderer {} \; but didn't find any struct definitions. I could only find TessResultRenderer as a class definition... https://github.com/tesseract-ocr/tesseract/blob/master/src/api/renderer.h#L45-L139 and the only thing that I guessed could possibly make sense was that C++ classes and structs could be used interchangeably. My google-fu failed to find anything useful, so an experiment... $ vi test.cpp #include <stdio.h> class SomeClass { public: int a; int b; }; typedef struct SomeClass SomeTypeDef; int main() { SomeTypeDef x; x.a = 5; x.b = 7; printf("Answer is %d\n", x.a + x.b); } $ gcc test.cpp $ ./a.out Answer is 12 Now I noticed that the TessResultRenderer member variables were private... https://github.com/tesseract-ocr/tesseract/blob/master/src/api/renderer.h#L131-L139 and curious about that I changed my test example from public to private which somewhat expectedly produced compile errors. So those TessResultRenderer member variables must only be accessed from a member function, but how is that C++ member function called from C to operate on a particular object? An example is TessResultRendererInsert... C Declaration: https://github.com/tesseract-ocr/tesseract/blob/c375f4fbf73b8f761b2e65e0e3ad6776b9fbee78/src/api/capi.h#L135 C Definition: https://github.com/tesseract-ocr/tesseract/blob/c375f4fbf73b8f761b2e65e0e3ad6776b9fbee78/src/api/capi.cpp#L90-L93 So in the C Defintion "the C++ member-function insert() as being called via a function pointer in the struct." (is that a reasonable way to describe it?) In this case, because of the private member variables, our FFI would treat TessResultRenderer as an opaque object, which simplifies things. I would guess in-Image direct access to the member variables from would need to account for the offset due to variables holding the function pointer to the member functions. cheers -ben P.S. for Tesseract FFI it might be good to start with reproducing this example... |
|
|
In reply to this post by Ben Coman
Can i just write a simple C shared library or DLL which calls the C++ ? So you are repackaging the C++ as a C library? I can’t see how this tack could fail to work. Just repackage C++ as C. You would have to come up with a procedural less OOP-ish API i guess. You could have C API functions F which take an Object as F’s first input and in this way each C++ Method becomes a C function. You only need wrap as much of the C++ API as you want to use and each C function just calls its C++ Method so making the wrappers is highly simple and mechanical i should think. it could even be automated. But i know some C++ but have never made anything in it. I suppose that if Smalltalk cannot contain a C++ Object then you could make a C struct which can be in Smalltalk and you have the API function copy this struct into the C++ Object then act on it then copy the Object data back into the struct which is in Smalltalk. But that’s a lot of work. Surely you can have a pointer to a C++ Object in Smalltalk. Maybe it would be better to have a separate C++ program P that you communicate with by sockets using Object handles H which are just Integer Array indexes into an Array of Objects in P? i suppose there could be a shared lib L that FFI could call which could call back program P if sockets were too slow or something. I guess Dolphin can input a Smalltalk BlockClosure B into an FFI call to L which could input B into program P which could call B to get back into Dolphin but i haven’t tried it myself. I guess there is a Smalltalk interface to Python via a socket and then from Python to C++ is easy? Seems like a code generator that has all this stuff figured out could be good. I think VisualWorks is probably good at connecting to C++ via FFI. What about chicken scheme or any of the C based Schemes? What about Smalltalk/X? borgLisp is an idea to make multiple Lisp dialects each isomorphic to its target language like C or C++ or Python or Ruby or Prolog or java or C# or Scheme or Rust etc any language can have an isomorphic Lisp dialect targeting it in order to bind all the languages into a single borgLisp where you can mix and match all the languages together. Where each Lisp dialect is just a simple Lisp code generator. And so once all the languages are in Lisp then all the Lisp things can be used to mix and match all the languages together and using Nix to set up and configure everything so everything works together one click like. all the different languages. so they can all work together in an easy generative format. So every language becomes Lisp and Lisp becomes every language. Using code generation you could even make a Debugger in Lisp and Smalltalk which could source debug any language like the Smalltalk debugger does for Smalltalk. but i guess this is off the topic On Fri, Nov 2, 2018 at 06:07 Ben Coman <[hidden email]> wrote:
|
|
|
Hello all, FFI was a little complexer than I had thought. And the Tesseract api was not helping either. But now I think I'm getting closer to make the example Ben proposed (https://github.com/tesseract-ocr/tesseract/wiki/APIExample, the C-program using the C-API) work in Squeak. Just one thing I cannot find an example for. I have to create the ExternalStructure classes for the structures PIXMAP and RGBA_QUAD. RGBA_QUAD is easy, but the PIXMAP-structure starts with an array of RGBA_QUADs. RGBA_QUAD[] does not seem to be working as a type specification, and RGBA_SQUAD* will reserve place for the first element, but not the whole array. Is there an example for such structures? From the Header file: 00101 struct PixColormap 00102 { 00103 void *array; /* colormap table (array of RGBA_QUAD) */ 00104 l_int32 depth; /* of pix (1, 2, 4 or 8 bpp) */ 00105 l_int32 nalloc; /* number of color entries allocated */ 00106 l_int32 n; /* number of color entries used */ 00107 }; 00108 typedef struct PixColormap PIXCMAP; 00109 00110 00111 /* Colormap table entry (after the BMP version). 00112 * Note that the BMP format stores the colormap table exactly 00113 * as it appears here, with color samples being stored sequentially, 00114 * in the order (b,g,r,a). */ 00115 struct RGBA_Quad 00116 { 00117 l_uint8 blue; 00118 l_uint8 green; 00119 l_uint8 red; 00120 l_uint8 reserved; 00121 }; 00122 typedef struct RGBA_Quad RGBA_QUAD; Op zo 4 nov. 2018 om 06:15 schreef Kjell Godo <[hidden email]>:
|
|
|
On Thu, 15 Nov 2018, Edwin Ancaer wrote:
> Hello all, > > FFI was a little complexer than I had thought. And the Tesseract api was not helping either. But now I think I'm getting closer to make the example Ben proposed > (https://github.com/tesseract-ocr/tesseract/wiki/APIExample, the C-program using the C-API) work in Squeak. > > Just one thing I cannot find an example for. I have to create the ExternalStructure classes for the structures PIXMAP and RGBA_QUAD. RGBA_QUAD is easy, but the PIXMAP-structure starts with an array of > RGBA_QUADs. RGBA_QUAD[] does not seem to be working as a type specification, and RGBA_SQUAD* will reserve place for the first element, but not the whole array. Is there an example for such > structures? RGBA_SQUAD* should reserve space for a pointer not for a whole struct. Also, if you don't need to access the contents of the struct directly and you don't have to allocate it yourself, then can just declare it as void*. Levente > > From the Header file: > > 00101 struct PixColormap > 00102 { > 00103 void *array; /* colormap table (array of RGBA_QUAD) */ > 00104 l_int32 depth; /* of pix (1, 2, 4 or 8 bpp) */ > 00105 l_int32 nalloc; /* number of color entries allocated */ > 00106 l_int32 n; /* number of color entries used */ > 00107 }; > 00108 typedef struct PixColormap PIXCMAP; > 00109 > 00110 > 00111 /* Colormap table entry (after the BMP version). > 00112 * Note that the BMP format stores the colormap table exactly > 00113 * as it appears here, with color samples being stored sequentially, > 00114 * in the order (b,g,r,a). */ > 00115 struct RGBA_Quad > 00116 { > 00117 l_uint8 blue; > 00118 l_uint8 green; > 00119 l_uint8 red; > 00120 l_uint8 reserved; > 00121 }; > 00122 typedef struct RGBA_Quad RGBA_QUAD; > > > Op zo 4 nov. 2018 om 06:15 schreef Kjell Godo <[hidden email]>: > Can i just write a simple C shared library or DLL which calls the C++ ? So you are repackaging the C++ as a C library? I can’t see how this tack could fail to work. Just repackage C++ as > C. > > You would have to come up with a procedural less OOP-ish API i guess. You could have C API functions F which take an Object as F’s first input and in this way each C++ Method becomes a C > function. You only need wrap as much of the C++ API as you want to use and each C function just calls its C++ Method so making the wrappers is highly simple and mechanical i should think. it > could even be automated. But i know some C++ but have never made anything in it. > > I suppose that if Smalltalk cannot contain a C++ Object then you could make a C struct which can be in Smalltalk and you have the API function copy this struct into the C++ Object then act on it > then copy the Object data back into the struct which is in Smalltalk. But that’s a lot of work. Surely you can have a pointer to a C++ Object in Smalltalk. > > Maybe it would be better to have a separate C++ program P that you communicate with by sockets using Object handles H which are just Integer Array indexes into an Array of Objects in P? i > suppose there could be a shared lib L that FFI could call which could call back program P if sockets were too slow or something. > > I guess Dolphin can input a Smalltalk BlockClosure B into an FFI call to L which could input B into program P which could call B to get back into Dolphin but i haven’t tried it myself. > > I guess there is a Smalltalk interface to Python via a socket and then from Python to C++ is easy? Seems like a code generator that has all this stuff figured out could be good. I think > VisualWorks is probably good at connecting to C++ via FFI. What about chicken scheme or any of the C based Schemes? What about Smalltalk/X? > > borgLisp is an idea to make multiple Lisp dialects each isomorphic to its target language like C or C++ or Python or Ruby or Prolog or java or C# or Scheme or Rust etc any language can have an > isomorphic Lisp dialect targeting it in order to bind all the languages into a single borgLisp where you can mix and match all the languages together. Where each Lisp dialect is just a simple > Lisp code generator. And so once all the languages are in Lisp then all the Lisp things can be used to mix and match all the languages together and using Nix to set up and configure everything > so everything works together one click like. all the different languages. so they can all work together in an easy generative format. So every language becomes Lisp and Lisp becomes every > language. Using code generation you could even make a Debugger in Lisp and Smalltalk which could source debug any language like the Smalltalk debugger does for Smalltalk. > > but i guess this is off the topic > > On Fri, Nov 2, 2018 at 06:07 Ben Coman <[hidden email]> wrote: > > > On Fri, 2 Nov 2018 at 18:44, Edwin Ancaer <[hidden email]> wrote: > Hello list, > > As I'm looking at a way to automate the search of documents in my humble administration, I read some articles about OCR. I came along an article about using Python with Tesseract, to > transform an scan of a document into text, that is searchable. > > My question now is if I can do something similar with Squeak. To my inexperienced eye, it seems like I should use FFI to call the functions in the Tesseract API, but this API is in > C++, and I don't know if it is possible to use FFI to call C++ functions? > > > You are right C++ is difficult because of the name mangling of function symbols, > but good fortune I notice Tesseract has C bindings... > https://github.com/tesseract-ocr/tesseract#for-developers > https://github.com/tesseract-ocr/tesseract/blob/master/src/api/capi.h > so it looks like you are in the clear. > > > Or should I forget the API and use OSProcess to start the tesseract program? > > > FFI will be more flexible. > > > Could anyone point me in the right direction, or just tell if the whole idea is insane? > > > I think its a great idea and actually Tesseract FFI is something I've wanted to play with before but not had the time. > I'd be interested to hear how you go with it. > > cheers -ben > > 1q > > > |
|
|
In reply to this post by Edwin Ancaer
On Thu, 15 Nov 2018 at 22:00, Edwin Ancaer <[hidden email]> wrote:
Cool!!!! Is Levente's advice sufficient? Or do you need more? Do you have a repo when your code is accessible? (and a few tests interested parties can run to familiarize with the system?) cheers -ben |
|
|
LEVENTE, Ben, RGBA_QUAD* is a pointer to the first element, of course, not the first element itself.Another blow for the ego...l By a repo, you mean on github? That also is strange territory, so I will have to ask for some patience. Thanks, Edwin |
|
|
In reply to this post by Edwin Ancaer
Not following the conversation fully, but a practical solution may be
to use a small Python program that calls tesseract and print things out to stdout. Then OSProcess can handle it.On Thu, Nov 15, 2018 at 6:01 AM Edwin Ancaer <[hidden email]> wrote: > > Hello all, > > FFI was a little complexer than I had thought. And the Tesseract api was not helping either. But now I think I'm getting closer to make the example Ben proposed (https://github.com/tesseract-ocr/tesseract/wiki/APIExample, the C-program using the C-API) work in Squeak. > > Just one thing I cannot find an example for. I have to create the ExternalStructure classes for the structures PIXMAP and RGBA_QUAD. RGBA_QUAD is easy, but the PIXMAP-structure starts with an array of RGBA_QUADs. RGBA_QUAD[] does not seem to be working as a type specification, and RGBA_SQUAD* will reserve place for the first element, but not the whole array. Is there an example for such structures? > > From the Header file: > > 00101 struct PixColormap > 00102 { > 00103 void *array; /* colormap table (array of RGBA_QUAD) */ > 00104 l_int32 depth; /* of pix (1, 2, 4 or 8 bpp) */ > 00105 l_int32 nalloc; /* number of color entries allocated */ > 00106 l_int32 n; /* number of color entries used */ > 00107 }; > 00108 typedef struct PixColormap PIXCMAP; > 00109 > 00110 > 00111 /* Colormap table entry (after the BMP version). > 00112 * Note that the BMP format stores the colormap table exactly > 00113 * as it appears here, with color samples being stored sequentially, > 00114 * in the order (b,g,r,a). */ > 00115 struct RGBA_Quad > 00116 { > 00117 l_uint8 blue; > 00118 l_uint8 green; > 00119 l_uint8 red; > 00120 l_uint8 reserved; > 00121 }; > 00122 typedef struct RGBA_Quad RGBA_QUAD; > > > > Op zo 4 nov. 2018 om 06:15 schreef Kjell Godo <[hidden email]>: >> >> Can i just write a simple C shared library or DLL which calls the C++ ? So you are repackaging the C++ as a C library? I can’t see how this tack could fail to work. Just repackage C++ as C. >> >> You would have to come up with a procedural less OOP-ish API i guess. You could have C API functions F which take an Object as F’s first input and in this way each C++ Method becomes a C function. You only need wrap as much of the C++ API as you want to use and each C function just calls its C++ Method so making the wrappers is highly simple and mechanical i should think. it could even be automated. But i know some C++ but have never made anything in it. >> >> I suppose that if Smalltalk cannot contain a C++ Object then you could make a C struct which can be in Smalltalk and you have the API function copy this struct into the C++ Object then act on it then copy the Object data back into the struct which is in Smalltalk. But that’s a lot of work. Surely you can have a pointer to a C++ Object in Smalltalk. >> >> Maybe it would be better to have a separate C++ program P that you communicate with by sockets using Object handles H which are just Integer Array indexes into an Array of Objects in P? i suppose there could be a shared lib L that FFI could call which could call back program P if sockets were too slow or something. >> >> I guess Dolphin can input a Smalltalk BlockClosure B into an FFI call to L which could input B into program P which could call B to get back into Dolphin but i haven’t tried it myself. >> >> I guess there is a Smalltalk interface to Python via a socket and then from Python to C++ is easy? Seems like a code generator that has all this stuff figured out could be good. I think VisualWorks is probably good at connecting to C++ via FFI. What about chicken scheme or any of the C based Schemes? What about Smalltalk/X? >> >> borgLisp is an idea to make multiple Lisp dialects each isomorphic to its target language like C or C++ or Python or Ruby or Prolog or java or C# or Scheme or Rust etc any language can have an isomorphic Lisp dialect targeting it in order to bind all the languages into a single borgLisp where you can mix and match all the languages together. Where each Lisp dialect is just a simple Lisp code generator. And so once all the languages are in Lisp then all the Lisp things can be used to mix and match all the languages together and using Nix to set up and configure everything so everything works together one click like. all the different languages. so they can all work together in an easy generative format. So every language becomes Lisp and Lisp becomes every language. Using code generation you could even make a Debugger in Lisp and Smalltalk which could source debug any language like the Smalltalk debugger does for Smalltalk. >> >> but i guess this is off the topic >> >> On Fri, Nov 2, 2018 at 06:07 Ben Coman <[hidden email]> wrote: >>> >>> >>> >>> On Fri, 2 Nov 2018 at 18:44, Edwin Ancaer <[hidden email]> wrote: >>>> >>>> Hello list, >>>> >>>> As I'm looking at a way to automate the search of documents in my humble administration, I read some articles about OCR. I came along an article about using Python with Tesseract, to transform an scan of a document into text, that is searchable. >>>> >>>> My question now is if I can do something similar with Squeak. To my inexperienced eye, it seems like I should use FFI to call the functions in the Tesseract API, but this API is in C++, and I don't know if it is possible to use FFI to call C++ functions? >>> >>> >>> You are right C++ is difficult because of the name mangling of function symbols, >>> but good fortune I notice Tesseract has C bindings... >>> https://github.com/tesseract-ocr/tesseract#for-developers >>> https://github.com/tesseract-ocr/tesseract/blob/master/src/api/capi.h >>> so it looks like you are in the clear. >>> >>> >>>> Or should I forget the API and use OSProcess to start the tesseract program? >>> >>> >>> FFI will be more flexible. >>> >>> >>>> Could anyone point me in the right direction, or just tell if the whole idea is insane? >>> >>> >>> I think its a great idea and actually Tesseract FFI is something I've wanted to play with before but not had the time. >>> I'd be interested to hear how you go with it. >>> >>> cheers -ben >>> >> 1q > > -- -- Yoshiki |
|
|
In reply to this post by Edwin Ancaer
On Fri, 16 Nov 2018 at 00:17, Edwin Ancaer <[hidden email]> wrote:
If not git, then www.squeaksource.com or ss3.gemtalksystems.com cheers -ben |
«
Return to Squeak - Dev
|
1 view|%1 views
| Free forum by Nabble | Edit this page |