You can improve text recognition accuracy by narrowing the range of characters valid for recognition. This way the recognition engine doesn't always have to choose its solutions from all 500 characters in the recognition engine's Total Character Set. (The multi-lingual omnifont MOR recognition module supports all of these characters; other recognition modules recognize fewer of them.) The Character Set concept is documented in detail in the Character Set topic. Broadly, the Set is compiled as follows:
- Language environment
This involves selecting one or more of the 114 available languages with the IG_REC_languages_set() function and optionally additional characters validated individually with the IG_REC_languagesplus_set function. Selecting only needed language(s) has a major impact. For example, selecting German only immediately INVALIDATES the Cyrillic and Greek alphabets and over 150 other unneeded accented letters. For a list of languages with their validated accented letters, see Languages and characters. For a list of accented letters and the languages which use them, see Characters, languages, modules. For an overview of the topic, see Introduction to language-related topics.
If IG_REC_languages_set() is not called, the default is to enable the English (Latin) alphabet. - Recognition module capabilities
Defining a recognition module for processing a zone may also restrict the available languages or characters within the Language environment.
- Filtering
The SDK provides filters (enumIGRecChrFilter) to further narrow down the Character Set by enabling only certain character classes, e.g., digits, uppercase letters, etc. The value IG_REC_FILTER_ALL means no filtering.
- Re-Expanding
An application may require exceptions to the filter rule. The most flexible way to re-expand the Character Set with individual characters after filtering is to specify them with a call to IG_REC_filterplus_set() and validate them with the IG_REC_FILTER_PLUS flag in the required zones.
- Zone-level modification
In addition to the global, page-level definition of the Character Set, the choice of Recognition module, filling method, filtering and use of the expansion string can be fine-tuned on a local, zone level. Auto-located zones will be set to take the global filter settings. Manually created or modified zones may contain zone-level settings. Local filtering and expansion (with IG_REC_FILTER_PLUS) can be set in the zone's filter field. The possible local filter values are the same as the global ones, with an extra one: IG_REC_FILTER_DEFAULT. If this is the only one set, the zone inherits the global filter setting.
The following examples illustrate various techniques for limiting the Character Set.
Recognition of a bi-lingual document (German and Spanish)
In this example, the default recognition module (omnifont, unless specifically changed) will be assigned to all zones, as will the filter value IG_REC_ FILTER_DEFAULT, i.e., there is no local modification of the Character Set.
| C |  Copy Code |
|---|---|
AT_ERRCOUNT nErrCount; AT_INT i; HIGEAR hIGear; HIG_REC_IMAGE hImg; enumIGRecLangEnable langs[IG_REC_LANG_SIZE]; for (i=0; i< IG_REC_LANG_SIZE; i++) { langs[i] = IG_REC_LANG_DISABLED; } langs[IG_REC_LANG_GER] = IG_REC_LANG_ENABLED; langs[IG_REC_LANG_SPA] = IG_REC_LANG_ENABLED; nErrCount = IG_REC_languages_set(langs); nErrCount = IG_load_file("Image.tif", &hIGear ); nErrCount = IG_REC_image_import(hIGear, &hImg); nErrCount = IG_image_delete(hIGear); nErrCount = IG_REC_image_recognize(hImg); //... nErrCount = IG_REC_image_delete(hImg); |
|
Character Set with no language selection: "a, A, b, B, c, C, d, D, e, E" as the only validated characters
This example illustrates a rare case, such as you have a page containing zones with a very restricted number of characters to be recognized, e.g., in recognizing forms or multiple-choice test papers. In this case the application doesn't enable any language, but instead defines the few characters necessary as LanguagesPlus characters. This means there is no language selection, and the Language environment consists solely of the individually defined LanguagesPlus characters. Also note that there is no filtering and no locally (zone) validated FilterPlus characters; in this case, the Language environment fully defines the Character Set, and it will be valid for the defined zone, and for others inserted with an identical zone structure.
In this case, the zone list is not empty, so the function IG_REC_image_recognize() will not perform auto-decomposition (auto-zoning), but will act on the inserted zone(s).
| C | Copy Code |
|---|---|
AT_ERRCOUNT nErrCount; HIGEAR hIGear; HIG_REC_IMAGE hImg; AT_INT i; enumIGRecLangEnable langs[IG_REC_LANG_SIZE]; for (i=0; i< IG_REC_LANG_SIZE; i++) { langs[i] = IG_REC_LANG_DISABLED; } nErrCount = IG_REC_languages_set(langs); // Specify the LanguagesPlus characters nErrCount = IG_REC_languagesplus_set(L"aAbBcCdDeE"); // nErrCount = IG_load_file("Image.tif", &hIGear ); nErrCount = IG_REC_image_import(hIGear, &hImg); nErrCount = IG_image_delete(hIGear); nErrCount = IG_REC_image_recognize(hImg); //... nErrCount = IG_REC_image_delete(hImg); |
|
Using language selection, LanguagesPlus characters, and local filter
The third example will read a printed page in Hungarian, in which a Croatian town name appears repeatedly, containing the character "z-hacek" in lower and uppercase. The Windows Eastern Europe Code Page (1250), is needed as the current Code Page (and for export). The page includes a table containing numbers, which should be zoned separately for digits-only recognition.
In this example, the Language environment is formed from the language selection (Hungarian) plus the two additional LanguagesPlus characters "z-hacek" and "Z-hacek". There is no global filter, but there is a local one, IG_REC_FILTER_DIGIT, defined for one zone.
| C | Copy Code |
|---|---|
AT_ERRCOUNT nErrCount; HIGEAR hIGear; HIG_REC_IMAGE hImg; AT_INT i; char c; WCHAR uni[256]; AT_REC_ZONE zone; AT_INT characterCount; enumIGRecLangEnable langs[IG_REC_LANG_SIZE]; IG_REC_output_codepage_set("Windows Eastern"); // Code Page 1250 for (i=0; i< IG_REC_LANG_SIZE; i++) { langs[i] = IG_REC_LANG_DISABLED; } langs[IG_REC_LANG_HUN] = IG_REC_LANG_ENABLED; nErrCount = IG_REC_languages_set(langs); memset(uni, 0, sizeof(uni)); //only one wide character terminating NULL is necessary i=0; c=0x9E; // Code of z-hacek in Code Page 1250 nErrCount = IG_REC_util_codepage_to_unicode(&c, 1, &uni[i++], &characterCount); c=0x8E; // Code of Z-hacek in Code Page 1250 nErrCount = IG_REC_util_codepage_to_unicode(&c, 1, &uni[i++], &characterCount); nErrCount = IG_REC_languagesplus_set(uni); nErrCount = IG_load_file("Image.tif", &hIGear ); nErrCount = IG_REC_image_import(hIGear, &hImg); nErrCount = IG_image_delete(hIGear); // 1st zone contains a table with digits memset(&zone, 0, sizeof(AT_REC_ZONE)); zone.Rect.left = 10; zone.Rect.top = 20; zone.Rect.right = 330; zone.Rect.bottom = 50; zone.FillingMethod = IG_REC_FM_OMNIFONT; zone.RecognitionModule = IG_REC_RM_OMNIFONT_MOR; zone.Filter = IG_REC_FILTER_DIGIT; zone.Type = IG_REC_WT_TABLE; nErrCount = IG_REC_zone_insert(hImg, 0, &zone); // 2nd zone contains flowed text without filtering memset(&zone, 0, sizeof(AT_REC_ZONE)); zone.Rect.left = 10; zone.Rect.top = 80; zone.Rect.right = 330; zone.Rect.bottom = 120; zone.FillingMethod = IG_REC_FM_OMNIFONT; zone.RecognitionModule = IG_REC_RM_OMNIFONT_MOR; zone.Type = IG_REC_WT_FLOW; zone.Filter = IG_REC_FILTER_ALL; nErrCount = IG_REC_zone_insert(hImg, 1, &zone); // 3rd zone contains flowed text without filtering // etc. nErrCount = IG_REC_image_recognize(hImg); //... nErrCount = IG_REC_image_delete(hImg); |
|