Add OCREngine parameter.

6 files changed, 97 insertions, 15 deletions

diff --git a/base/tessocr.cpp b/base/tessocr.cpp
index 26e5432c7..057a4a369 100644
--- a/base/tessocr.cpp
+++ b/base/tessocr.cpp
@@ -186,9 +186,10 @@ tess_file_reader(const char *fname, GenericVector<char> *out)
 }
 
 int
-ocr_init_api(gs_memory_t *mem, const char *language, void **state)
+ocr_init_api(gs_memory_t *mem, const char *language, int engine, void **state)
 {
     tesseract::TessBaseAPI *api;
+    enum tesseract::OcrEngineMode mode;
 
     leptonica_mem = mem->non_gc_memory;
     setPixMemoryManager(my_leptonica_malloc, my_leptonica_free);
@@ -202,10 +203,28 @@ ocr_init_api(gs_memory_t *mem, const char *language, void **state)
         return_error(gs_error_VMerror);
     }
 
+    switch (engine)
+    {
+        case OCR_ENGINE_DEFAULT:
+            mode = tesseract::OcrEngineMode::OEM_DEFAULT;
+            break;
+        case OCR_ENGINE_LSTM:
+            mode = tesseract::OcrEngineMode::OEM_LSTM_ONLY;
+            break;
+        case OCR_ENGINE_LEGACY:
+            mode = tesseract::OcrEngineMode::OEM_TESSERACT_ONLY;
+            break;
+        case OCR_ENGINE_BOTH:
+            mode = tesseract::OcrEngineMode::OEM_TESSERACT_LSTM_COMBINED;
+            break;
+        default:
+            return_error(gs_error_rangecheck);
+    }
+
     // Initialize tesseract-ocr with English, without specifying tessdata path
     if (api->Init(NULL, 0, /* data, data_size */
                   language,
-                  tesseract::OcrEngineMode::OEM_DEFAULT,
+                  mode,
                   NULL, 0, /* configs, configs_size */
                   NULL, NULL, /* vars_vec */
                   false, /* set_only_non_debug_params */
@@ -265,8 +284,7 @@ do_ocr_image(gs_memory_t *mem,
              int w, int h, int bpp, int raster,
              int xres, int yres, void *data, int restore,
              int hocr, int pagecount,
-             const char *language,
-             char **out)
+             const char *language, int engine, char **out)
 {
     char *outText;
     tesseract::TessBaseAPI *api;
@@ -277,7 +295,7 @@ do_ocr_image(gs_memory_t *mem,
 
     if (language == NULL || *language == 0)
         language = "eng";
-    code = ocr_init_api(mem, language, (void **)&api);
+    code = ocr_init_api(mem, language, engine, (void **)&api);
     if (code < 0)
         return code;
 
@@ -328,19 +346,20 @@ do_ocr_image(gs_memory_t *mem,
 int ocr_image_to_hocr(gs_memory_t *mem,
                       int w, int h, int bpp, int raster,
                       int xres, int yres, void *data, int restore,
-                      int pagecount, const char *language, char **out)
+                      int pagecount, const char *language,
+                      int engine, char **out)
 {
         return do_ocr_image(mem, w, h, bpp, raster, xres, yres, data,
-                            restore, 1, pagecount, language, out);
+                            restore, 1, pagecount, language, engine, out);
 }
 
 int ocr_image_to_utf8(gs_memory_t *mem,
                       int w, int h, int bpp, int raster,
                       int xres, int yres, void *data, int restore,
-                      const char *language, char **out)
+                      const char *language, int engine, char **out)
 {
         return do_ocr_image(mem, w, h, bpp, raster, xres, yres, data,
-                            restore, 0, 0, language, out);
+                            restore, 0, 0, language, engine, out);
 }
 
 int
diff --git a/base/tessocr.h b/base/tessocr.h
index c5e7967b9..78c30a0d4 100644
--- a/base/tessocr.h
+++ b/base/tessocr.h
@@ -21,18 +21,27 @@
 
 #include "gsmemory.h"
 
+enum
+{
+    OCR_ENGINE_DEFAULT = 0,
+    OCR_ENGINE_LSTM = 1,
+    OCR_ENGINE_LEGACY = 2,
+    OCR_ENGINE_BOTH = 3
+};
+
 int ocr_image_to_utf8(gs_memory_t *mem,
                       int w, int h, int bpp, int raster,
                       int xres, int yres,
                       void *data, int restore_data,
-                      const char *language, char **out);
+                      const char *language, int engine, char **out);
 
 int ocr_image_to_hocr(gs_memory_t *mem,
                       int w, int h, int bpp, int raster,
                       int xres, int yres, void *data, int restore,
-                      int pagecount, const char *language, char **out);
+                      int pagecount, const char *language,
+                      int engine, char **out);
 
-int ocr_init_api(gs_memory_t *mem, const char *language, void **state);
+int ocr_init_api(gs_memory_t *mem, const char *language, int engine, void **state);
 
 void ocr_fin_api(gs_memory_t *mem, void *api_);
 
diff --git a/devices/gdevocr.c b/devices/gdevocr.c
index ddfaa809e..44cc72341 100644
--- a/devices/gdevocr.c
+++ b/devices/gdevocr.c
@@ -43,6 +43,7 @@ struct gx_device_ocr_s {
     gx_prn_device_common;
     gx_downscaler_params downscale;
     char language[1024];
+    int engine;
     int page_count;
 };
 
@@ -127,6 +128,9 @@ ocr_get_params(gx_device * dev, gs_param_list * plist)
     if ((code = param_write_string(plist, "OCRLanguage", &langstr)) < 0)
         ecode = code;
 
+    if ((code = param_write_string(plist, "OCREngine", &pdev->engine)) < 0)
+        ecode = code;
+
     if ((code = gx_downscaler_write_params(plist, &pdev->downscale,
                                            GX_DOWNSCALER_PARAMS_MFS)) < 0)
         ecode = code;
@@ -146,6 +150,7 @@ ocr_put_params(gx_device *dev, gs_param_list *plist)
     gs_param_string langstr;
     const char *param_name;
     size_t len;
+    int engine;
 
     switch (code = param_read_string(plist, (param_name = "OCRLanguage"), &langstr)) {
         case 0:
@@ -162,6 +167,17 @@ ocr_put_params(gx_device *dev, gs_param_list *plist)
             param_signal_error(plist, param_name, ecode);
     }
 
+    switch (code = param_read_int(plist, (param_name = "OCREngine"), &engine)) {
+        case 0:
+            pdev->engine = engine;
+            break;
+        case 1:
+            break;
+        default:
+            ecode = code;
+            param_signal_error(plist, param_name, ecode);
+    }
+
     code = gx_downscaler_read_params(plist, &pdev->downscale,
                                      GX_DOWNSCALER_PARAMS_MFS);
     if (code < 0)
@@ -225,14 +241,14 @@ do_ocr_print_page(gx_device_ocr * pdev, gp_file * file, int hocr)
                                  (int)pdev->HWResolution[0],
                                  (int)pdev->HWResolution[1],
                                  data, 0, pdev->page_count,
-                                 "eng", &out);
+                                 "eng", pdev->engine, &out);
     else
         code = ocr_image_to_utf8(pdev->memory,
                                  width, height,
                                  8, raster,
                                  (int)pdev->HWResolution[0],
                                  (int)pdev->HWResolution[1],
-                                 data, 0, "eng", &out);
+                                 data, 0, "eng", pdev->engine, &out);
     if (code < 0)
         goto done;
     if (out)
diff --git a/devices/gdevpdfimg.h b/devices/gdevpdfimg.h
index baf8175e0..3fe7aac22 100644
--- a/devices/gdevpdfimg.h
+++ b/devices/gdevpdfimg.h
@@ -64,6 +64,7 @@ typedef struct gx_device_pdf_image_s {
     /* OCR data */
     struct {
         char language[1024];
+        int engine;
         void *state;
 
         /* Number of "file level" objects - i.e. the number of objects
diff --git a/devices/gdevpdfocr.c b/devices/gdevpdfocr.c
index 95d358b18..b03116945 100644
--- a/devices/gdevpdfocr.c
+++ b/devices/gdevpdfocr.c
@@ -63,6 +63,7 @@ pdfocr_put_some_params(gx_device * dev, gs_param_list * plist)
     gs_param_string langstr;
     const char *param_name;
     size_t len;
+    int engine;
 
     switch (code = param_read_string(plist, (param_name = "OCRLanguage"), &langstr)) {
         case 0:
@@ -79,6 +80,17 @@ pdfocr_put_some_params(gx_device * dev, gs_param_list * plist)
             param_signal_error(plist, param_name, ecode);
     }
 
+    switch (code = param_read_int(plist, (param_name = "OCREngine"), &engine)) {
+        case 0:
+            pdf_dev->ocr.engine = engine;
+            break;
+        case 1:
+            break;
+        default:
+            ecode = code;
+            param_signal_error(plist, param_name, ecode);
+    }
+
     return code;
 }
 
@@ -120,6 +132,9 @@ pdfocr_get_some_params(gx_device * dev, gs_param_list * plist)
     if ((code = param_write_string(plist, "OCRLanguage", &langstr)) < 0)
         ecode = code;
 
+    if ((code = param_write_int(plist, "OCREngine", &pdf_dev->ocr.engine)) < 0)
+        ecode = code;
+
     return ecode;
 }
 
@@ -416,7 +431,7 @@ ocr_file_init(gx_device_pdf_image *dev)
     stream_write(dev->strm, funky_font6a, sizeof(funky_font6a));
     stream_write(dev->strm, funky_font6b, sizeof(funky_font6b)-1);
 
-    return ocr_init_api(dev->memory, language, &dev->ocr.state);
+    return ocr_init_api(dev->memory, language, dev->ocr.engine, &dev->ocr.state);
 }
 
 static void
diff --git a/doc/Devices.htm b/doc/Devices.htm
index 1a64c92a0..d6a656af3 100644
--- a/doc/Devices.htm
+++ b/doc/Devices.htm
@@ -999,6 +999,28 @@ resolution independence, and editability.</p>
 </blockquote>
 </blockquote>
 <p>
+  The system is designed to allow different OCR engines to be used.
+  Even with Tesseract, there are 2 different engines, the &quot;legacy&quot;
+  engine, and the &quot;LSTM&quot; engine. These have different tradeoffs
+  regarding speed, accuracy, sensitivity to different fonts, size of
+  data etc. The engines in use can be changed by using the
+  <code>-dOCREngine=</code> switch;
+</p>
+<blockquote>
+<dl>
+<dt><code>-dOCREngine=</code><b><em>integer</em></b></dt>
+<dd>This sets the engine to use for OCR. It is the callers responsibility
+  to ensure that the traineddata supplied is appropriate for the selected
+  engine.
+  <dl>
+    <dt>0<dd>Default engine.
+    <dt>1<dd>LSTM engine only.
+    <dt>2<dd>Tesseract &quot;legacy&quot; engine only.
+    <dt>3<dd>Legacy and LSTM Engine combined.
+  </dl>
+</dd></dl>
+</blockquote>
+<p>
   The first device is named ocr. It extracts data as unicode codepoints
   and outputs them to the device as a stream of UTF-8 bytes.
 </p>