ALOCR Ver 1.0 reference manual

1. Constants and Structures
   
   1. Character Control
   2. Recognition Result
   
   
2. Details
   
   1. Types of Pattern Dictionaries
   2. Construction of Pattern Dictionary Classes
   3. User Dictionary Management
   4. Dictionary Class:CJocrDict Reference
   5. Construction of Pattern Classes
   6. Feature Calculation of Patterns
   7. Pattern Class:CJocrPattern Reference
   8. Construction of Single Character Recognition Classes
      Updated on November 6, 2000 In addition to the license code, the license code file can now be specified.
   9. Single Character Recognition
   10. Single Character Recognition Class:CJocrRecognize Reference
   11. Construction of Line Recognition Classes
   12. Line Recognition
   13. Line Recognition Class:CJocrLine Reference
   14. Language Processing with Line Recognition Class:CJocrLang Reference
   15. Paragraph Recognition Class:CJocrBlock Reference
       Updated on July 31, 2002 Skew can now be specified for paragraphs. It is possible to recognize italic text only for single-line text.
       Updated on August 31, 2005 Addition of new dictionaries
       Updated on August 31, 2005 New version recognition method. Significant improvement in recognition rate compared to the conventional method
       Updated on November 30, 2022 Deep learning support. Support for color image and grayscale image recognition. Significant improvement in recognition rate compared to the conventional method
       
   
   
   
3. Sample Programs
   
   1. VC++ Sample Project
   
   

// Pattern bounding rectangle
typedef struct {
    short       x1;         // Top left x coordinate of the pattern
    short       y1;         // Top left y coordinate of the pattern
    short       x2;         // Bottom right x coordinate of the pattern
    short       y2;         // Bottom right y coordinate of the pattern
} OCRRect;
// Recognition candidate
typedef struct {
    char*           code;       // Pointer to the pattern string
    unsigned char   score;      // Confidence level
    unsigned char   filler[3];  // padding
} Candidate;
//////////////////
// OCR result structure
// 128 bytes fixed
typedef struct {
    Candidate       cand[MAX_CAND]; // Candidate data MAX_CAND == 10 80 bytes
    OCRRect         area;           // Recognized area 8 bytes
    long            fieldtype;
    unsigned long   chartype;       // Character type of the recognition result
    unsigned long   maskchartype1;
    unsigned long   maskchartype2;
    unsigned long   space;          // Number of spaces following the characters in a line recognition
    unsigned char   cgravx;
    unsigned char   cgravy;
    unsigned char   morph;
    unsigned char   size;
    char            newcand[KEYSIZE_MAX];
} OCRResult;

typedef struct {
	short 		x1;			// Top left x-coordinate of the pattern
	short 		y1;			// Top left y-coordinate of the pattern
	short 		x2;			// Bottom right x-coordinate of the pattern
	short 		y2;			// Bottom right y-coordinate of the pattern
} OCRRect;

1. Basic Dictionary Format (2)
2. Optional Dictionary Format (any number)
3. User Dictionary Format (0 or 1)

1. Creation of an instance
2. Setting of the basic dictionary name, optional dictionary name, and user dictionary name
3. Loading of the dictionary

Code example for constructing a basic dictionary class

#include	"ocrdef.h"
#include	"ocrco.h"
#include	"cjocrstock.h"
#include	"cjocrdict98.h"
#include	"errcode.h"

.....

// 1...Create an instance of the dictionary class
  	CJocrDict* pjocrdict = new CJocrDict;
// 2...Set up the basic dictionary
	pjocrdict->msetsystemdict("\\dic\\feature\\system");
	pjocrdict->msetsystemdict("\\dic\\feature\\systemfat");
// 3...Set up the optional dictionary
	pjocrdict->msetdiffdict("\\dic\\feature\\diff1");
	pjocrdict->msetdiffdict("\\dic\\feature\\diff2");
	pjocrdict->msetdiffdict("\\dic\\feature\\diff3");
	pjocrdict->msetdiffdict("\\dic\\feature\\diff4");
	pjocrdict->msetdiffdict("\\dic\\feature\\diff5");
// 5...Set up the user dictionary (optional)
	m_pJocrDict->msetuserdict("\\dic\\feature\\userpat");
// 6...Load the dictionary
	i1 = pjocrdict->mloaddict();
	if(i1 < 0) {
		Error (defined in errcode.h)
		FILE_OPEN_ERROR	The dictionary could not be found
		FILE_READ_ERROR	The dictionary could not be read
		FILE_SEEK_ERROR	Unable to seek to the dictionary record
		MEMORY_SHORTAGE	Out of memory (no further processing)
		FATAL_ERROR		Unrecoverable error (no further processing)
	}

......Recognition, registration, deletion, reference

// Delete the instance
	delete pjocrdict;

1. Seek and register
   Assuming pattern is an instance of the CJocrPattern class, after setting the image of the kanji "漢" as the pattern and calculating its features (calling pattern->mcalcall(...)),
   

   	int i1 = pjocrdict->mput("漢",pattern);
   	if(i1 < 0) {
   		Error;
   		FILE_SEEK_ERROR
   		FILE_WRITE_ERROR
   	}
   

2. Seek and delete
   To delete a record corresponding to the key "漢" from the user dictionary, first seek and then call the deletion function as follows.
   

   	int i1 = pjocrdict->mseek("漢");
   	if(i1 < 0) {
   		Error;
   	}
   	if(i1 == 1) {
   		// Delete
   		i1 = pjocrdict->mdel();
   		if(i1 < 0) {
   			Error;
   		}
   	}
   

3. Seek to the next record and get the key and the image of the pattern.
   

   	int i1 = pjocrdict->mseeknext();
   	// keysize is a variable used for input and output
   	// It represents the maximum read buffer size when inputting
   	// and the actual read size when outputting
   	unsigned long	keysize = KEYSIZE_MAX;
   	char			keybuffer[KEYSIZE_MAX];
   	if(i1 == 1) {
   		i1 = pjocrdict->mgetkey(keysize,keybuffer);
   		if(i1 < 0) {
   			Error;
   		}
   		else {
   			// REG_FONT_SIZE is the normalized font size in bytes
   			// It is resized and saved as REG_FONT_SIZE during user dictionary registration
   			// In ALOCR Ver.1.0, the normalized font is 48x48 pixels, which is 288 bytes
   			// REG_FONT_WIDTH....48
   			// REG_FONT_HEIGHT....48
   			// REG_FONT_SIZE....48*48/8(8 pixels = 1 byte)
   			// recordsize is a variable used for input and output
   			// It represents the maximum read buffer size when inputting
   			// and the actual read size when outputting
   			unsigned long	recordsize  = REG_FONT_SIZE;
   			char			record[REG_FONT_SIZE];
   			i1 = mgetpattern(recordsize,record);
   			if(i1 < 0) {
   				Error;
   			}
   			// The obtained pattern is a bitmap of REG_FONT_WIDTH × REG_FONT_HEIGHT pixels.
   		}
   	}
   	else if(i1 == 0) {
   		No more records;
   	}
   	else {
   		Error;
   	}
   

1. Setting up the base dictionary
   
2. Setting up the option dictionary (difference dictionary)
   
3. Setting up the user dictionary
   
4. Loading the set dictionaries
   
5. Seeking
   
6. Getting the pattern registration key
   
7. Getting the pattern image
   
8. Registration
   
9. Deletion
   
10. Getting the number of records
    
11. Changing the upper limit of user dictionary registrations
    

• Two to three-digit numbers or unit symbols in vertical writing
• Cases where one pattern becomes more than 3 characters, such as "TEL" or "(株)"
• Cases where the designation of a single pattern, such as a logo on a business card, cannot be expressed in one character

	// 1...Create an instance of the pattern class
	CJocrPattern* pattern = new CJocrPattern;
	// 2...Allocate memory
	i1 = pattern->mallocmemory();
	if(i1 < 0) {Display error message; delete pattern;}

	.....

	delete pattern;

typedef struct {
	unsigned char* 	top;		// Address of the start of the image data
	short			width;		// Width of the image data (in byte boundaries, in pixels)
	short			height;		// Height of the image data (in pixels)
} OCRBuffer;
typedef struct {
	short		x1;			// Top-left x-coordinate of the pattern
	short		y1;			// Top-left y-coordinate of the pattern
	short		x2;			// Bottom-right x-coordinate of the pattern
	short		y2;			// Bottom-right y-coordinate of the pattern
} OCRRect;

1. Feature Calculation
   

    // 1...Create an instance of the recognition class
    // Construct an instance by specifying a 20-digit code supplied by the library license distributor.

    // Updated on November 6, 2000
    // Alternatively, specify the path to a license code file supplied by the library license distributor.

    CJocrRecognize* precognize = new CJocrRecognize("ABCDEFGHJKLMNPQ23456");
    // For license code files CJocrRecognize* precognize = new CJocrRecognize("C:\\Program Files\\Foo\\jocr.kcd");
    // 2...Pattern setting
    // Let pattern be an instance of the CJocrPattern class
    // .....Generate and initialize the pattern
    precognize->msetpatter(pattern);
    // 3...Dictionary setting
    // Let pjocrdict be an instance of the CJocrDict class
    // .....Generate and initialize pjocrdict
    precognize->msetdict(pjocrdict);
    // 4...Memory allocation
    i1 = precognize->mallocmemory();
    if(i1 < 0) {
        MEMORY_SHORTAGE....Insufficient memory (defined in errcode.h)
        Display error message;delete pattern;
    }
    ....
    Repeat single character recognition process
    ....
    delete pattern;

• CHAR_SET_ALL All types of characters
• CHAR_SET_ALPHABET Alphabet characters
• CHAR_SET_NUMBER Numbers

    pattern->mrecognize(CHAR_SET_ALL);
    // Get recognition result
    // For OCRResult, please refer to section 2-2 in the reference manual.
    OCRResult aresult;
    mgetresult(&aresult);

1. Initialization
   
2. Recognition, Identification, and Retrieval of Recognition Results
   

	// 1...Create an instance of the line class
  	CJocrLine* pjocrline = new CJocrLine;
	// 2...Set an instance of the pattern class to the line class (pattern already constructed elsewhere)
	//     Execution of mallocmemory is required
	pjocrline->msetpattern(pattern);
	// 3...Set an instance of the recognition class to the line class (precognize already constructed elsewhere)
	//     Execution of msetpattern, msetdict, and mallocmemory is required
	pjocrline->msetrecognize(precognize);
	// 4...Initialize the document
	// Call this whenever the image buffer for recognition changes
	typedef struct {
		unsigned char* 	top;		// Starting address of the image data
		short			width;		// Width of the image data (byte alignment, pixel units)
		short			height;		// Height of the image data (pixel units)
	} OCRBuffer;
	OCRBuffer aocrbuffer;
	aocrbuffer.top = ...;		// Buffer address
	aocrbuffer.width = ...;		// Buffer width (pixel units, multiple of 8)
	aocrbuffer.height = ...;	// Buffer height (pixel units)
	int i1 = pjocrline->msetdocument(&aocrbuffer);
	if(i1 < 0) {
		// MEMORY_SHORTAGE....Insufficient memory
		Display error message;
		delete pjocrline;
	}
    pjocrline->msetdpi(400);    // Resolution set to 400dpi
	Repeat line recognition. Call msetdocument when the image buffer for recognition changes.

	delete pjocrline;

    // Line settings
    OCRRect aocrrect;
    aocrrect.x1;   // Top-left X coordinate of the bounding rectangle of the line
    aocrrect.y1;   // Top-left Y coordinate of the bounding rectangle of the line
    aocrrect.x2;   // Bottom-right X coordinate of the bounding rectangle of the line
    aocrrect.y2;   // Bottom-right Y coordinate of the bounding rectangle of the line
#if Horizontal writing
    msetlineuser(&aocrrect,HORIZONTAL_LINE); // Horizontal writing
#else
    msetlineuser(&aocrrect,VERTICAL_LINE);   // Vertical writing
#endif
    // Line recognition
    i1 = pjocrline->mrecognize(CHAR_SET_ALL);
    if(i1 < 0) {Display error message;}
    else {
        int         resultnum;
        OCRResult   pocrresult[resultnum];
        // Get results
        i1 = pjocrline->mgetresult(resultnum,pocrresult);
        if(i1 < 0) {Display error message;}
    }

1. Initialization
   
2. Document Settings
   
3. Line Settings, Line Recognition, and Retrieving Recognition Results
   

1. Initialization
   

1. Initialization
   
2. Setting up the document
   Skew of the paragraph can be specified. It is possible to recognize italicized text limited to a single line.
   
3. Setting up the paragraph, creating and obtaining normalized images
   
4. Paragraph recognition, obtaining recognition result
   
5. New version recognition method
   
6. Deep learning compatible recognition method