![maXbox Starter 103 ¨C Text recognition of characters in images.
Image to Text API
8
Blaise Pascal Magazine 109 2022
Article Page 1/14
maXbox
maXbox
function Image_to_text_API2(AURL, url_imgpath, aApikey: string): string;
var httpq: THttpConnectionWinInet;
rets: TStringStream;
heads: TStrings; iht:IHttpConnection2;
begin
httpq:= THttpConnectionWinInet.Create(true);
rets:= TStringStream.create('');
heads:= TStringlist.create;
try
heads.add('apikey='+aAPIkey);
iht:= httpq.setHeaders(heads);
httpq.Get(Format(AURL,[url_imgpath]),rets);
if httpq.getresponsecode=200 Then result:= rets.datastring
else result:='Failed:'+
itoa(Httpq.getresponsecode)+Httpq.GetResponseHeader('message');
except
writeln('EWI_HTTP: '+ExceptiontoString(exceptiontype,exceptionparam));
finally
httpq:= Nil;
heads.Free;
rets.Free;
end;
end;
The main part function opens a connection, invokes the API and results a stream which we convert to a
datastring.
Image2Text or Image to Text live demo is providing an API service on its APILayer publication platform.
Live Demo feature allows you to test the API within your browser; no need to install or code anything.
You can modify all the parameters as you like and interact with the API from many languages. The API
export format is JSON, e.g. our book cover see below:
maXbox](https://image.slidesharecdn.com/blaiseuk109maxkleinerimage2textapi-230411103435-dd26a2e3/85/Blaise_UK_109_Max-Kleiner_image2textAPI-pdf-2-320.jpg)
![9
Blaise Pascal Magazine 109 2022
Article Page 1/14
maXbox
maXbox
The published result datasets are based on LSTM in combination with a OCR. LSTM
stands for Long Short-Term Memory and is a type of Recurrent Neural Network(RNN).
Talking about RNN, it is a network that works on the present input by taking into
consideration the previous output (feedback) and storing in its memory as memory cells
for a short period of time (short-term memory). For example take our book-cover as input:
LSTMs have feedback connections and cells which
make them different to more traditional feed-forward
neural networks with the still existing vanishing gradient
problem.
This property enables LSTMs to process entire sequences
of data (e.g. time series, handwriting or sentences) without
treating each point in the sequence independently,
but rather, retaining useful information about previous
data in the sequence like "Objektorientiert", "modellieren",
"und", "entwickeln" as a context.
The output of the call
writeln(Image_to_text_API2(URL_APILAY,
URLIMAGEPATH4,
DNwCF9Rf6y1AmSSednjn8ZhAxYXr----'));
is the JSON datastring in about Runtime:
0:0:3.859:
{"lang": "de", "all_text":
"ih u00bb Der EntwicklernFachwissen
fu00fcr Programmierer
nMax KleinernUMLnmit DelphinObjektorientiert modellieren
nund entwickelnnSoftware & Support", "annotations": ["ih", "u00bb",
"Der", "Entwickler", "Fachwissen", "fu00fcr", "Programmierer", "Max",
"Kleiner", "UML", "mit", "Delphi", "Objektorientiert", "modellieren",
"und", "entwickeln", "Software", "&", "Support"]}
maXbox Starter 103 ¨C Text recognition of characters in images.
Image to Text API
maXbox
Also the well known Tesseract 4.0 (like OmniPage) added a new OCR engine based on
LSTM neural networks.
The API can also be triggered with this few lines of P4D code:
procedure PyCode(imgpath: string);
begin
with TPythonEngine.Create(Nil) do begin
pythonhome:= 'C:UsersmaxAppDataLocalProgramsPythonPython36-32';
try
loadDLL;
ExecString('import requests');
ExecStr('url= "https://api.apilayer.com/image_to_text/url?
url='+imgpath+'"');
ExecStr('payload = {}');
ExecStr('headers= {"apikey": "dy5L70eQx72794XBZ8sewEgYTZR85----"}');
Println(EvalStr('requests.request("GET",url, headers=headers,
data=payload).text'));
except
raiseError;
finally
unloadDLL;
free;
end;
end;
end;](https://image.slidesharecdn.com/blaiseuk109maxkleinerimage2textapi-230411103435-dd26a2e3/85/Blaise_UK_109_Max-Kleiner_image2textAPI-pdf-3-320.jpg)
![Our final example is an interesting one. What about an old
wine bottle (1993) with shapes (and grapes ?)), an old historic
painting from 1619, dates, symbols and position angles:
https://my6code.files.wordpress.com/2022/12/
wine_1993_20221230_141947.jpg?w=768
10
Blaise Pascal Magazine 109 2022
Article Page 1/14
maXbox
maXbox
When you fail with a restricted call or an
invalid key you get a bunch of exceptions like the following:
maXbox Starter 103 ¨C Text recognition of characters in images.
Image to Text API
maXbox
function GetWinInetError(ErrorCode:Cardinal): string;
const
winetdll = 'wininet.dll';
var
Len : Integer;
Buffer: PChar;
begin
Len := FormatMessage(
FORMAT_MESSAGE_FROM_HMODULE or FORMAT_MESSAGE_FROM_SYSTEM or
FORMAT_MESSAGE_ALLOCATE_BUFFER or FORMAT_MESSAGE_IGNORE_INSERTS or
FORMAT_MESSAGE_ARGUMENT_ARRAY,
Pointer(GetModuleHandle(winetdll)), ErrorCode,0, @Buffer,SizeOf(Buffer),
nil);
try
while (Len > 0) and {$IFDEF UNICODE}(CharInSet(Buffer[Len - 1],[#0.
.#32,
'.'])) {$ELSE}(Buffer[Len- 1] in [#0..#32, '.']) {$ENDIF} do
Dec(Len);
SetString(Result, Buffer, Len);
finally
LocalFree(HLOCAL(Buffer));
end;
end;
winininet_error: Unauthorized (401). or
{"message":"Invalid authentication credentials"}
The fact that error code is not "one of the expected return values" tells for the versions
that the error comes from an underlying layer and this API just passes it up on internal
failure.To shine a bit more light on those errors a function exists to convert the ErrorCode to
a string for better understanding:
And the result is convincing, also the fact that the year
in the label image was recognized correctly as 1619:
{"lang": "it", "all_text":
"DSnPODERE CAPACCIAn1619nLuggo deffo apacia
nQUERCIAGRANDEnVino da tavola della Toscanan1993
nProdotto e imbottigliato all'originenPodere Capaccia
nPacini Giampaolo & C. s.a.s.nRadda in Chianti (SI) - ItalianItalia",
"annotations": ["DS", "PODERE", "CAPACCIA", "1619", "Luggo", "deffo",
"apacia", "QUERCIAGRANDE", "Vino", "da", "tavola", "della", "Toscana",
"1993", "Prodotto", "e", "imbottigliato", "all'origine", "Podere", "Capaccia",
"Pacini", "Giampaolo", "&", "C.", "s.a.s.", "Radda", "in", "Chianti", "(", "SI", ")", "-", "Italia", "Italia"]}](https://image.slidesharecdn.com/blaiseuk109maxkleinerimage2textapi-230411103435-dd26a2e3/85/Blaise_UK_109_Max-Kleiner_image2textAPI-pdf-4-320.jpg)
Blaise_UK_109_Max Kleiner_image2textAPI.pdf
- 1. 7 Blaise Pascal Magazine 109 2022 Starter Expert maXbox maXbox maXbox Starter 103 ¨C Text recognition of characters in images. Image to Text API Article Page 1/14 ¡°A picture is worth a thousand words. An interface is worth a thousand pictures.¡±. This register allows you a monthly usage of 30 successful calls. Almost all API's has a free plan to subscribe. Looking at the following book-cover, it will extract the text information easily, even though the cover has shadows and positioned with angle. We use WinHttp.WinHttpRequest, JSONObjects and TGraphics library with loading and testing the REST-client. Also we pass the API-key as a request-header, so get a key first at: https://apilayer.com/marketplace You can also use a powerful 'OCR' feature (text in picture recognition) to extract text from an image during the conversion process. In this case, you will get an editable text document as a result that you can adjust and modify as you need. The data represents is JSON data with all the text extracted and even the language of the text to scan is auto detected. Before we dive into code this is the main part of the script: This API recognizes and reads a text embedded in pictures or photos. Image to Text API uses a neural net (LSTM) based OCR engine which is focused on line recognition, but also supports recognizing the character patterns. It supports both handwriting and printed materials as well as street maps. APILayer is an API marketplace where also your API can reach a broader audiences, but first you need an API-key for free: The result of a simple subscription will be the screenshot below: maXbox
- 2. maXbox Starter 103 ¨C Text recognition of characters in images. Image to Text API 8 Blaise Pascal Magazine 109 2022 Article Page 1/14 maXbox maXbox function Image_to_text_API2(AURL, url_imgpath, aApikey: string): string; var httpq: THttpConnectionWinInet; rets: TStringStream; heads: TStrings; iht:IHttpConnection2; begin httpq:= THttpConnectionWinInet.Create(true); rets:= TStringStream.create(''); heads:= TStringlist.create; try heads.add('apikey='+aAPIkey); iht:= httpq.setHeaders(heads); httpq.Get(Format(AURL,[url_imgpath]),rets); if httpq.getresponsecode=200 Then result:= rets.datastring else result:='Failed:'+ itoa(Httpq.getresponsecode)+Httpq.GetResponseHeader('message'); except writeln('EWI_HTTP: '+ExceptiontoString(exceptiontype,exceptionparam)); finally httpq:= Nil; heads.Free; rets.Free; end; end; The main part function opens a connection, invokes the API and results a stream which we convert to a datastring. Image2Text or Image to Text live demo is providing an API service on its APILayer publication platform. Live Demo feature allows you to test the API within your browser; no need to install or code anything. You can modify all the parameters as you like and interact with the API from many languages. The API export format is JSON, e.g. our book cover see below: maXbox
- 3. 9 Blaise Pascal Magazine 109 2022 Article Page 1/14 maXbox maXbox The published result datasets are based on LSTM in combination with a OCR. LSTM stands for Long Short-Term Memory and is a type of Recurrent Neural Network(RNN). Talking about RNN, it is a network that works on the present input by taking into consideration the previous output (feedback) and storing in its memory as memory cells for a short period of time (short-term memory). For example take our book-cover as input: LSTMs have feedback connections and cells which make them different to more traditional feed-forward neural networks with the still existing vanishing gradient problem. This property enables LSTMs to process entire sequences of data (e.g. time series, handwriting or sentences) without treating each point in the sequence independently, but rather, retaining useful information about previous data in the sequence like "Objektorientiert", "modellieren", "und", "entwickeln" as a context. The output of the call writeln(Image_to_text_API2(URL_APILAY, URLIMAGEPATH4, DNwCF9Rf6y1AmSSednjn8ZhAxYXr----')); is the JSON datastring in about Runtime: 0:0:3.859: {"lang": "de", "all_text": "ih u00bb Der EntwicklernFachwissen fu00fcr Programmierer nMax KleinernUMLnmit DelphinObjektorientiert modellieren nund entwickelnnSoftware & Support", "annotations": ["ih", "u00bb", "Der", "Entwickler", "Fachwissen", "fu00fcr", "Programmierer", "Max", "Kleiner", "UML", "mit", "Delphi", "Objektorientiert", "modellieren", "und", "entwickeln", "Software", "&", "Support"]} maXbox Starter 103 ¨C Text recognition of characters in images. Image to Text API maXbox Also the well known Tesseract 4.0 (like OmniPage) added a new OCR engine based on LSTM neural networks. The API can also be triggered with this few lines of P4D code: procedure PyCode(imgpath: string); begin with TPythonEngine.Create(Nil) do begin pythonhome:= 'C:UsersmaxAppDataLocalProgramsPythonPython36-32'; try loadDLL; ExecString('import requests'); ExecStr('url= "https://api.apilayer.com/image_to_text/url? url='+imgpath+'"'); ExecStr('payload = {}'); ExecStr('headers= {"apikey": "dy5L70eQx72794XBZ8sewEgYTZR85----"}'); Println(EvalStr('requests.request("GET",url, headers=headers, data=payload).text')); except raiseError; finally unloadDLL; free; end; end; end;
- 4. Our final example is an interesting one. What about an old wine bottle (1993) with shapes (and grapes ?)), an old historic painting from 1619, dates, symbols and position angles: https://my6code.files.wordpress.com/2022/12/ wine_1993_20221230_141947.jpg?w=768 10 Blaise Pascal Magazine 109 2022 Article Page 1/14 maXbox maXbox When you fail with a restricted call or an invalid key you get a bunch of exceptions like the following: maXbox Starter 103 ¨C Text recognition of characters in images. Image to Text API maXbox function GetWinInetError(ErrorCode:Cardinal): string; const winetdll = 'wininet.dll'; var Len : Integer; Buffer: PChar; begin Len := FormatMessage( FORMAT_MESSAGE_FROM_HMODULE or FORMAT_MESSAGE_FROM_SYSTEM or FORMAT_MESSAGE_ALLOCATE_BUFFER or FORMAT_MESSAGE_IGNORE_INSERTS or FORMAT_MESSAGE_ARGUMENT_ARRAY, Pointer(GetModuleHandle(winetdll)), ErrorCode,0, @Buffer,SizeOf(Buffer), nil); try while (Len > 0) and {$IFDEF UNICODE}(CharInSet(Buffer[Len - 1],[#0. .#32, '.'])) {$ELSE}(Buffer[Len- 1] in [#0..#32, '.']) {$ENDIF} do Dec(Len); SetString(Result, Buffer, Len); finally LocalFree(HLOCAL(Buffer)); end; end; winininet_error: Unauthorized (401). or {"message":"Invalid authentication credentials"} The fact that error code is not "one of the expected return values" tells for the versions that the error comes from an underlying layer and this API just passes it up on internal failure.To shine a bit more light on those errors a function exists to convert the ErrorCode to a string for better understanding: And the result is convincing, also the fact that the year in the label image was recognized correctly as 1619: {"lang": "it", "all_text": "DSnPODERE CAPACCIAn1619nLuggo deffo apacia nQUERCIAGRANDEnVino da tavola della Toscanan1993 nProdotto e imbottigliato all'originenPodere Capaccia nPacini Giampaolo & C. s.a.s.nRadda in Chianti (SI) - ItalianItalia", "annotations": ["DS", "PODERE", "CAPACCIA", "1619", "Luggo", "deffo", "apacia", "QUERCIAGRANDE", "Vino", "da", "tavola", "della", "Toscana", "1993", "Prodotto", "e", "imbottigliato", "all'origine", "Podere", "Capaccia", "Pacini", "Giampaolo", "&", "C.", "s.a.s.", "Radda", "in", "Chianti", "(", "SI", ")", "-", "Italia", "Italia"]}
- 5. 11 Blaise Pascal Magazine 109 2022 Article Page 1/14 maXbox maXbox maXbox Starter 103 ¨C Text recognition of characters in images. Image to Text API maXbox Any missing or incomplete data is difficult to find without knowing the original. But on the other side, it is very easy to use since users just need to screenshot the part they wish to convert and then copy the text after. Furthermore the API access is provided in a REST-like interface (Representational State Transfer) exposing database resources or pre-trained models in a JSON format with content-type in the Response Header. NOTE: If a programming language is not listed in the Code Example from the live demo, you can still make API calls by using a HTTP request library written in our programming language, as we did with GET or POST.
- 6. 12 Blaise Pascal Magazine 109 2022 Article Page 1/14 maXbox maXbox Conclusion: The Image to Text API from APILayer detects and extracts text from images using state-of-the-art optical character recognition (OCR) algorithms in combination with a neural network called LSTM. It can detect texts of different sizes, fonts, and even handwriting or difficult numbers. Reference: https://apilayer.com/marketplace/image_to_text-api https://apilayer.com/docs https://my6.code.blog/2022/09/02/webpostdata/ http://www.kleiner.ch/kleiner/images/uml_buch.jpg Doc and Tool: https://maxbox4.wordpress.com Script Ref: 1176_APILayer_Demo1.txt Appendix: A Delphi REST client API to consume REST services written in any programming language with a class from maXbox4 integration: https://github.com/fabriciocolombo/delphi-rest-client-api/blob/master/ src/HttpConnectionWinInet.pas The API it is designed to work with Delphi 7 or later; newer versions takes advantage of Generics Methods. https://github.com/fabriciocolombo/delphi-rest-client-api maXbox maXbox Starter 103 ¨C Text recognition of characters in images. Image to Text API Saturday 15th April 2023