Merge pull request #365 from duanyao/covered_text_handling

Covered text handling

CMakeLists.txt

@@ -161,6 +161,10 @@ set(PDF2HTMLEX_SRC ${PDF2HTMLEX_SRC}
     src/Base64Stream.cc
     src/Color.h
     src/Color.cc
+    src/CoveredTextDetector.h
+    src/CoveredTextDetector.cc
+    src/DrawingTracer.h
+    src/DrawingTracer.cc
     src/HTMLState.h
     src/HTMLTextLine.h
     src/HTMLTextLine.cc

pdf2htmlEX.1.in

@@ -242,6 +242,11 @@ If set to 0, pdf2htmlEX would try its best to balance the two methods above.
 .B \-\-optimize\-text <0|1> (Default: 0)
 If set to 1, pdf2htmlEX will try to reduce the number of HTML elements used for text. Turn it off if anything goes wrong.
 
+.TP
+.B --correct-text-visibility <0|1> (Default: 0)
+If set to 1, pdf2htmlEX will try to detect texts covered by other graphics and properly arrange them, 
+i.e. covered texts are made transparent in text layer, and are drawn on background layer.
+
 .SS Background Image
 
 .TP

src/BackgroundRenderer/CairoBackgroundRenderer.cc

@@ -63,6 +63,13 @@ void CairoBackgroundRenderer::drawChar(GfxState *state, double x, double y,
     {
         CairoOutputDev::drawChar(state,x,y,dx,dy,originX,originY,code,nBytes,u,uLen);
     }
+    // If a char is treated as image, it is not subject to cover test
+    // (see HTMLRenderer::drawString), so don't increase drawn_char_count.
+    else if (param.correct_text_visibility) {
+        if (html_renderer->is_char_covered(drawn_char_count))
+            CairoOutputDev::drawChar(state,x,y,dx,dy,originX,originY,code,nBytes,u,uLen);
+        drawn_char_count++;
+    }
 }
 
 void CairoBackgroundRenderer::beginTextObject(GfxState *state)
@@ -104,6 +111,7 @@ static GBool annot_cb(Annot *, void * pflag) {
 
 bool CairoBackgroundRenderer::render_page(PDFDoc * doc, int pageno)
 {
+    drawn_char_count = 0;
     double page_width;
     double page_height;
     if(param.use_cropbox)

src/BackgroundRenderer/CairoBackgroundRenderer.h

@@ -67,6 +67,7 @@ private:
   std::unordered_map<int, int> bitmaps_ref_count;
   // id of bitmaps' stream used by current page
   std::vector<int> bitmaps_in_current_page;
+  int drawn_char_count;
 };
 
 }

src/BackgroundRenderer/SplashBackgroundRenderer.cc

@@ -88,6 +88,13 @@ void SplashBackgroundRenderer::drawChar(GfxState *state, double x, double y,
     {
         SplashOutputDev::drawChar(state,x,y,dx,dy,originX,originY,code,nBytes,u,uLen);
     }
+    // If a char is treated as image, it is not subject to cover test
+    // (see HTMLRenderer::drawString), so don't increase drawn_char_count.
+    else if (param.correct_text_visibility) {
+        if (html_renderer->is_char_covered(drawn_char_count))
+            SplashOutputDev::drawChar(state,x,y,dx,dy,originX,originY,code,nBytes,u,uLen);
+        drawn_char_count++;
+    }
 }
 
 void SplashBackgroundRenderer::beginTextObject(GfxState *state)
@@ -129,6 +136,7 @@ static GBool annot_cb(Annot *, void * pflag) {
 
 bool SplashBackgroundRenderer::render_page(PDFDoc * doc, int pageno)
 {
+    drawn_char_count = 0;
     bool process_annotation = param.process_annotation;
     doc->displayPage(this, pageno, param.h_dpi, param.v_dpi,
             0, 

src/BackgroundRenderer/SplashBackgroundRenderer.h

@@ -71,6 +71,7 @@ protected:
   HTMLRenderer * html_renderer;
   const Param & param;
   std::string format;
+  int drawn_char_count;
 };
 
 } // namespace pdf2htmlEX

src/CoveredTextDetector.cc

@@ -0,0 +1,51 @@
+/*
+ * CoveredTextDetector.cc
+ *
+ *  Created on: 2014-6-14
+ *      Author: duanyao
+ */
+
+#include "CoveredTextDetector.h"
+
+#include "util/math.h"
+
+namespace pdf2htmlEX {
+
+void CoveredTextDetector::reset()
+{
+    char_bboxes.clear();
+    chars_covered.clear();
+}
+
+void CoveredTextDetector::add_char_bbox(double * bbox)
+{
+    char_bboxes.insert(char_bboxes.end(), bbox, bbox + 4);
+    chars_covered.push_back(false);
+}
+
+void CoveredTextDetector::add_char_bbox_clipped(double * bbox, bool patially)
+{
+    char_bboxes.insert(char_bboxes.end(), bbox, bbox + 4);
+    chars_covered.push_back(true);
+    if (patially)
+        add_non_char_bbox(bbox, chars_covered.size() - 1);
+}
+
+void CoveredTextDetector::add_non_char_bbox(double * bbox, int index)
+{
+    if (index < 0)
+        index = chars_covered.size();
+    for (int i = 0; i < index; i++)
+    {
+        if (chars_covered[i])
+            continue;
+        double * cbbox = &char_bboxes[i * 4];
+        if (bbox_intersect(cbbox, bbox))
+        {
+            chars_covered[i] = true;
+            add_non_char_bbox(cbbox, i);
+        }
+    }
+}
+
+}

src/CoveredTextDetector.h

@@ -0,0 +1,61 @@
+/*
+ * CoveredTextDetector.h
+ *
+ *  Created on: 2014-6-14
+ *      Author: duanyao
+ */
+
+#ifndef COVEREDTEXTDETECTOR_H__
+#define COVEREDTEXTDETECTOR_H__
+
+#include <vector>
+
+namespace pdf2htmlEX {
+
+/**
+ * Detect characters that are covered by non-char graphics on a page.
+ */
+class CoveredTextDetector
+{
+public:
+
+    /**
+     * Reset to initial state. Should be called when start drawing a page.
+     */
+    void reset();
+
+    /**
+     * Add a drawn character's bounding box.
+     * @param bbox (x0, y0, x1, y1)
+     */
+    void add_char_bbox(double * bbox);
+
+    void add_char_bbox_clipped(double * bbox, bool patially);
+
+    /**
+     * Add a drawn non-char graphics' bounding box.
+     * If it intersects any previously drawn char's bbox, the char is marked as covered
+     * and treated as an non-char.
+     * @param bbox (x0, y0, x1, y1)
+     * @param index this graphics' drawing order: assume it is drawn after (index-1)th
+     *   char. -1 means after the last char.
+     */
+    void add_non_char_bbox(double * bbox, int index = -1);
+
+    /**
+     * An array of flags indicating whether a char is covered by any non-char graphics.
+     * Index by the order that these chars are added.
+     * This vector grows as add_char_bbox() is called, so its size is the count
+     * of currently drawn chars.
+     */
+    const std::vector<bool> & get_chars_covered() { return chars_covered; }
+
+private:
+    std::vector<bool> chars_covered;
+    // x00, y00, x01, y01; x10, y10, x11, y11;...
+    std::vector<double> char_bboxes;
+};
+
+}
+
+#endif /* COVEREDTEXTDETECTOR_H__ */

src/DrawingTracer.cc

@@ -0,0 +1,360 @@
+/*
+ * DrawingTracer.cc
+ *
+ *  Created on: 2014-6-15
+ *      Author: duanyao
+ */
+
+#include "GfxFont.h"
+
+#include "util/math.h"
+#include "DrawingTracer.h"
+
+//#define DT_DEBUG(x)  (x)
+#define DT_DEBUG(x)
+
+#if !ENABLE_SVG
+#warning "Cairo is disabled because ENABLE_SVG is off, --correct-text-visibility has limited functionality."
+#endif
+
+namespace pdf2htmlEX
+{
+
+DrawingTracer::DrawingTracer(const Param & param): param(param)
+#if ENABLE_SVG
+, cairo(nullptr)
+#endif
+{
+}
+
+DrawingTracer::~DrawingTracer()
+{
+    finish();
+}
+
+void DrawingTracer::reset(GfxState *state)
+{
+    if (!param.correct_text_visibility)
+        return;
+    finish();
+
+#if ENABLE_SVG
+    cairo_rectangle_t page_box {0, 0, width:state->getPageWidth(), height:state->getPageHeight()};
+    cairo_surface_t * surface = cairo_recording_surface_create(CAIRO_CONTENT_COLOR_ALPHA, &page_box);
+    cairo = cairo_create(surface);
+#endif
+}
+
+void DrawingTracer::finish()
+{
+#if ENABLE_SVG
+    if (cairo)
+    {
+        cairo_destroy(cairo);
+        cairo = nullptr;
+    }
+#endif
+}
+
+// Poppler won't inform us its initial CTM, and the initial CTM is affected by zoom level.
+// OutputDev::clip() may be called before OutputDev::updateCTM(), so we can't rely on GfxState::getCTM(),
+// and should trace ctm changes ourself (via cairo).
+void DrawingTracer::update_ctm(GfxState *state, double m11, double m12, double m21, double m22, double m31, double m32)
+{
+    if (!param.correct_text_visibility)
+        return;
+
+#if ENABLE_SVG
+    cairo_matrix_t matrix;
+    matrix.xx = m11;
+    matrix.yx = m12;
+    matrix.xy = m21;
+    matrix.yy = m22;
+    matrix.x0 = m31;
+    matrix.y0 = m32;
+    cairo_transform(cairo, &matrix);
+#endif
+}
+
+void DrawingTracer::clip(GfxState * state, bool even_odd)
+{
+    if (!param.correct_text_visibility)
+        return;
+#if ENABLE_SVG
+    do_path(state, state->getPath());
+    cairo_set_fill_rule(cairo, even_odd? CAIRO_FILL_RULE_EVEN_ODD : CAIRO_FILL_RULE_WINDING);
+    cairo_clip (cairo);
+#endif
+}
+
+void DrawingTracer::clip_to_stroke_path(GfxState * state)
+{
+    if (!param.correct_text_visibility)
+        return;
+    // TODO cairo_stroke_to_path() ?
+}
+
+void DrawingTracer::save()
+{
+    if (!param.correct_text_visibility)
+        return;
+#if ENABLE_SVG
+    cairo_save(cairo);
+#endif
+}
+void DrawingTracer::restore()
+{
+    if (!param.correct_text_visibility)
+        return;
+#if ENABLE_SVG
+    cairo_restore(cairo);
+#endif
+}
+
+void DrawingTracer::do_path(GfxState * state, GfxPath * path)
+{
+#if ENABLE_SVG
+    //copy from CairoOutputDev::doPath
+    GfxSubpath *subpath;
+    int i, j;
+    double x, y;
+    cairo_new_path(cairo);
+    for (i = 0; i < path->getNumSubpaths(); ++i) {
+        subpath = path->getSubpath(i);
+        if (subpath->getNumPoints() > 0) {
+            x = subpath->getX(0);
+            y = subpath->getY(0);
+            cairo_move_to(cairo, x, y);
+            j = 1;
+            while (j < subpath->getNumPoints()) {
+                if (subpath->getCurve(j)) {
+                    x = subpath->getX(j+2);
+                    y = subpath->getY(j+2);
+                    cairo_curve_to(cairo,
+                        subpath->getX(j), subpath->getY(j),
+                        subpath->getX(j+1), subpath->getY(j+1),
+                        x, y);
+                    j += 3;
+                } else {
+                    x = subpath->getX(j);
+                    y = subpath->getY(j);
+                    cairo_line_to(cairo, x, y);
+                    ++j;
+                }
+            }
+            if (subpath->isClosed()) {
+                cairo_close_path (cairo);
+            }
+        }
+    }
+#endif
+}
+
+void DrawingTracer::stroke(GfxState * state)
+{
+#if ENABLE_SVG
+    if (!param.correct_text_visibility)
+        return;
+
+    DT_DEBUG(printf("DrawingTracer::stroke\n"));
+
+    cairo_set_line_width(cairo, state->getLineWidth());
+
+    // GfxPath is broken into steps, each step makes up a cairo path and its bbox is used for covering test.
+    // TODO
+    // 1. path steps that are not vertical or horizontal lines may still falsely "cover" many chars,
+    // can we slice those steps further?
+    // 2. if the line width is small, can we just ignore the path?
+    // 3. line join feature can't be retained. We use line-cap-square to minimize the problem that
+    //   some chars actually covered by a line join are missed. However chars covered by a acute angle
+    //   with line-join-miter may be still recognized as not covered.
+    cairo_set_line_cap(cairo, CAIRO_LINE_CAP_SQUARE);
+    GfxPath * path = state->getPath();
+    for (int i = 0; i < path->getNumSubpaths(); ++i) {
+        GfxSubpath * subpath = path->getSubpath(i);
+        if (subpath->getNumPoints() <= 0)
+            continue;
+        double x = subpath->getX(0);
+        double y = subpath->getY(0);
+        //p: loop cursor; j: next point index
+        int p =1, j = 1;
+        int n = subpath->getNumPoints();
+        while (p <= n) {
+            cairo_new_path(cairo);
+            cairo_move_to(cairo, x, y);
+            if (subpath->getCurve(j)) {
+                x = subpath->getX(j+2);
+                y = subpath->getY(j+2);
+                cairo_curve_to(cairo,
+                    subpath->getX(j), subpath->getY(j),
+                    subpath->getX(j+1), subpath->getY(j+1),
+                    x, y);
+                p += 3;
+            } else {
+                x = subpath->getX(j);
+                y = subpath->getY(j);
+                cairo_line_to(cairo, x, y);
+                ++p;
+            }
+
+            DT_DEBUG(printf("DrawingTracer::stroke:new box:\n"));
+            double sbox[4];
+            cairo_stroke_extents(cairo, sbox, sbox + 1, sbox + 2, sbox + 3);
+            if (sbox[0] != sbox[2] && sbox[1] != sbox[3])
+                draw_non_char_bbox(state, sbox);
+            else
+                DT_DEBUG(printf("DrawingTracer::stroke:zero box!\n"));
+
+            if (p == n)
+            {
+                if (subpath->isClosed())
+                    j = 0; // if sub path is closed, go back to starting point
+                else
+                    break;
+            }
+            else
+                j = p;
+        }
+    }
+#endif
+}
+
+void DrawingTracer::fill(GfxState * state, bool even_odd)
+{
+    if (!param.correct_text_visibility)
+        return;
+
+#if ENABLE_SVG
+    do_path(state, state->getPath());
+    //cairo_fill_extents don't take fill rule into account.
+    //cairo_set_fill_rule (cairo, even_odd? CAIRO_FILL_RULE_EVEN_ODD : CAIRO_FILL_RULE_WINDING);
+    double fbox[4];
+    cairo_fill_extents(cairo, fbox, fbox + 1, fbox + 2, fbox + 3);
+    draw_non_char_bbox(state, fbox);
+#endif
+}
+
+void DrawingTracer::draw_non_char_bbox(GfxState * state, double * bbox)
+{
+#if ENABLE_SVG
+    double cbox[4];
+    cairo_clip_extents(cairo, cbox, cbox + 1, cbox + 2, cbox + 3);
+    if(bbox_intersect(cbox, bbox, bbox))
+#endif
+    {
+        transform_bbox_by_ctm(bbox, state);
+        DT_DEBUG(printf("DrawingTracer::draw_non_char_bbox:[%f,%f,%f,%f]\n", bbox[0],bbox[1],bbox[2],bbox[3]));
+        if (on_non_char_drawn)
+            on_non_char_drawn(bbox);
+    }
+}
+
+void DrawingTracer::draw_char_bbox(GfxState * state, double * bbox)
+{
+#if ENABLE_SVG
+    // Note: even if 4 corners of the char are all in or all out of the clip area,
+    // it could still be partially clipped.
+    // TODO better solution?
+    int pt_in = 0;
+    if (cairo_in_clip(cairo, bbox[0], bbox[1]))
+        ++pt_in;
+    if (cairo_in_clip(cairo, bbox[2], bbox[3]))
+        ++pt_in;
+    if (cairo_in_clip(cairo, bbox[2], bbox[1]))
+        ++pt_in;
+    if (cairo_in_clip(cairo, bbox[0], bbox[3]))
+        ++pt_in;
+
+    if (pt_in == 0)
+    {
+        transform_bbox_by_ctm(bbox);
+        if(on_char_clipped)
+            on_char_clipped(bbox, false);
+    }
+    else
+    {
+        if (pt_in < 4)
+        {
+            double cbox[4];
+            cairo_clip_extents(cairo, cbox, cbox + 1, cbox + 2, cbox + 3);
+            bbox_intersect(cbox, bbox, bbox);
+        }
+        transform_bbox_by_ctm(bbox);
+        if (pt_in < 4)
+        {
+            if(on_char_clipped)
+                on_char_clipped(bbox, true);
+        }
+        else
+        {
+            if (on_char_drawn)
+                on_char_drawn(bbox);
+        }
+    }
+#else
+    transform_bbox_by_ctm(bbox, state);
+    if (on_char_drawn)
+        on_char_drawn(bbox);
+#endif
+    DT_DEBUG(printf("DrawingTracer::draw_char_bbox:[%f,%f,%f,%f]\n",bbox[0],bbox[1],bbox[2],bbox[3]));
+}
+
+void DrawingTracer::draw_image(GfxState *state)
+{
+    if (!param.correct_text_visibility)
+        return;
+    double bbox[4] {0, 0, 1, 1};
+    draw_non_char_bbox(state, bbox);
+}
+
+void DrawingTracer::draw_char(GfxState *state, double x, double y, double ax, double ay)
+{
+    if (!param.correct_text_visibility)
+        return;
+
+    Matrix tm, itm;
+    memcpy(tm.m, state->getTextMat(), sizeof(tm.m));
+
+    double cx = state->getCurX(), cy = state->getCurY(), fs = state->getFontSize(),
+            ry = state->getRise(), h = state->getHorizScaling();
+
+    //cx and cy has been transformed by text matrix, we need to reverse them.
+    tm.invertTo(&itm);
+    double char_cx, char_cy;
+    itm.transform(cx, cy, &char_cx, &char_cy);
+
+    //TODO Vertical? Currently vertical/type3 chars are treated as non-chars.
+    double char_m[6] {fs * h, 0, 0, fs, char_cx + x, char_cy + y + ry};
+
+    double final_m[6];
+    tm_multiply(final_m, tm.m, char_m);
+
+    auto font = state->getFont();
+    double bbox[4] {0, 0, ax, ay};
+    double desc = font->getDescent(), asc = font->getAscent();
+    if (font->getWMode() == 0)
+    {
+        bbox[1] += desc;
+        bbox[3] += asc;
+    }
+    else
+    {//TODO Vertical?
+    }
+    tm_transform_bbox(final_m, bbox);
+    draw_char_bbox(state, bbox);
+}
+
+
+void DrawingTracer::transform_bbox_by_ctm(double * bbox, GfxState * state)
+{
+#if ENABLE_SVG
+    cairo_matrix_t mat;
+    cairo_get_matrix(cairo, &mat);
+    double mat_a[6] {mat.xx, mat.yx, mat.xy, mat.yy, mat.x0, mat.y0};
+    tm_transform_bbox(mat_a, bbox);
+#else
+    tm_transform_bbox(state->getCTM(), bbox);
+#endif
+}
+
+} /* namespace pdf2htmlEX */

src/DrawingTracer.h

@@ -0,0 +1,79 @@
+/*
+ * DrawingTracer.h
+ *
+ *  Created on: 2014-6-15
+ *      Author: duanyao
+ */
+
+#ifndef DRAWINGTRACER_H__
+#define DRAWINGTRACER_H__
+
+#include <functional>
+
+#include <GfxState.h>
+
+#include "pdf2htmlEX-config.h"
+
+#if ENABLE_SVG
+#include <cairo.h>
+#endif
+
+#include "Param.h"
+
+namespace pdf2htmlEX
+{
+
+class DrawingTracer
+{
+public:
+    /*
+     * The callback to receive drawn event.
+     * bbox in device space.
+     */
+    // a non-char graphics is drawn
+    std::function<void(double * bbox)> on_non_char_drawn;
+    // a char is drawn in the clip area
+    std::function<void(double * bbox)> on_char_drawn;
+    // a char is drawn out of/partially in the clip area
+    std::function<void(double * bbox, bool patially)> on_char_clipped;
+
+    DrawingTracer(const Param & param);
+    virtual ~DrawingTracer();
+    void reset(GfxState * state);
+
+    /*
+     * A character is drawing
+     * x, y: glyph-drawing position, in PDF text object space.
+     * ax, ay: glyph advance, in glyph space.
+     */
+    void draw_char(GfxState * state, double x, double y, double ax, double ay);
+    /*
+     * An image is drawing
+     */
+    void draw_image(GfxState * state);
+    void update_ctm(GfxState * state, double m11, double m12, double m21, double m22, double m31, double m32);
+    void clip(GfxState * state, bool even_odd = false);
+    void clip_to_stroke_path(GfxState * state);
+    void fill(GfxState * state, bool even_odd = false);
+    void stroke(GfxState * state);
+    void save();
+    void restore();
+
+private:
+    void finish();
+    // Following methods operate in user space (just before CTM is applied)
+    void do_path(GfxState * state, GfxPath * path);
+    void draw_non_char_bbox(GfxState * state, double * bbox);
+    void draw_char_bbox(GfxState * state, double * bbox);
+    // If cairo is available, parameter state is ignored
+    void transform_bbox_by_ctm(double * bbox, GfxState * state = nullptr);
+
+    const Param & param;
+
+#if ENABLE_SVG
+    cairo_t * cairo;
+#endif
+};
+
+} /* namespace pdf2htmlEX */
+#endif /* DRAWINGTRACER_H__ */

src/HTMLRenderer/HTMLRenderer.h

@@ -31,10 +31,13 @@
 #include "HTMLTextPage.h"
 
 #include "BackgroundRenderer/BackgroundRenderer.h"
+#include "CoveredTextDetector.h"
+#include "DrawingTracer.h"
 
 #include "util/const.h"
 #include "util/misc.h"
 
+
 namespace pdf2htmlEX {
 
 class HTMLRenderer : public OutputDev
@@ -89,7 +92,9 @@ public:
      * We just mark as changed, and recheck if they have been changed when we are about to output a new string
      */
 
-    virtual void restoreState(GfxState * state) { updateAll(state); }
+    virtual void restoreState(GfxState * state);
+
+    virtual void saveState(GfxState *state);
 
     virtual void updateAll(GfxState * state);
 
@@ -125,15 +130,34 @@ public:
 
     virtual void drawImage(GfxState * state, Object * ref, Stream * str, int width, int height, GfxImageColorMap * colorMap, GBool interpolate, int *maskColors, GBool inlineImg);
 
-    virtual void stroke(GfxState *state) { css_do_path(state, false); }
-    virtual void fill(GfxState *state) { css_do_path(state, true); }
+    virtual void drawSoftMaskedImage(GfxState *state, Object *ref, Stream *str,
+                       int width, int height,
+                       GfxImageColorMap *colorMap,
+                       GBool interpolate,
+                       Stream *maskStr,
+                       int maskWidth, int maskHeight,
+                       GfxImageColorMap *maskColorMap,
+                       GBool maskInterpolate);
+
+    virtual void stroke(GfxState *state); ////{ css_do_path(state, false); }
+    virtual void fill(GfxState *state); ////{ css_do_path(state, true); }
+    virtual void eoFill(GfxState *state);
     virtual GBool axialShadedFill(GfxState *state, GfxAxialShading *shading, double tMin, double tMax);
 
     virtual void processLink(AnnotLink * al);
 
     /* capacity test */
-    bool can_stroke(GfxState *state) { return css_do_path(state, false, true); }
-    bool can_fill(GfxState *state) { return css_do_path(state, true, true); }
+    bool can_stroke(GfxState *state) { return false; } ////{ return css_do_path(state, false, true); }
+    bool can_fill(GfxState *state) { return false; } ////{ return css_do_path(state, true, true); }
+
+    /*
+     * Covered text handling.
+     */
+    // Is a char (actually a glyph) covered by non-char's. Index in drawing order in current page.
+    // Does not fail on out-of-bound conditions, but return false.
+    bool is_char_covered(int index);
+    // Currently drawn char (glyph) count in current page.
+    int get_char_count() { return (int)covered_text_detecor.get_chars_covered().size(); }
 
 protected:
     ////////////////////////////////////////////////////
@@ -195,6 +219,7 @@ protected:
     // make sure the current HTML style consistent with PDF
     void prepare_text_line(GfxState * state);
 
+#if 0 //disable CSS drawing
     ////////////////////////////////////////////////////
     // CSS drawing
     ////////////////////////////////////////////////////
@@ -214,6 +239,7 @@ protected:
             double * line_width_array, int line_width_count,
             const GfxRGB * line_color, const GfxRGB * fill_color, 
             void (*style_function)(void *, std::ostream &) = nullptr, void * style_function_data = nullptr );
+#endif //disable CSS drawing
 
 
     ////////////////////////////////////////////////////
@@ -328,7 +354,6 @@ protected:
 #endif
     BackgroundRenderer * bg_renderer;
     BackgroundRenderer * fallback_bg_renderer;
-    bool fallback_bg_required;
 
     struct {
         std::ofstream fs;
@@ -338,6 +363,9 @@ protected:
     std::string cur_page_filename;
 
     static const std::string MANIFEST_FILENAME;
+
+    CoveredTextDetector covered_text_detecor;
+    DrawingTracer tracer;
 };
 
 } //namespace pdf2htmlEX

src/HTMLRenderer/draw.cc

@@ -30,6 +30,39 @@ using std::sqrt;
 using std::vector;
 using std::ostream;
 
+void HTMLRenderer::restoreState(GfxState * state)
+{
+    updateAll(state);
+    tracer.restore();
+}
+
+void HTMLRenderer::saveState(GfxState *state)
+{
+    tracer.save();
+}
+
+void HTMLRenderer::stroke(GfxState * state)
+{
+    tracer.stroke(state);
+}
+
+void HTMLRenderer::fill(GfxState * state)
+{
+    tracer.fill(state);
+}
+
+void HTMLRenderer::eoFill(GfxState * state)
+{
+    tracer.fill(state, true);
+}
+
+GBool HTMLRenderer::axialShadedFill(GfxState *state, GfxAxialShading *shading, double tMin, double tMax)
+{
+    tracer.fill(state); //TODO correct?
+    return true;
+}
+
+#if 0 //disable css drawing
 static bool is_horizontal_line(GfxSubpath * path)
 {
     return ((path->getNumPoints() == 2)
@@ -415,6 +448,7 @@ void HTMLRenderer::css_draw_rectangle(double x, double y, double w, double h, co
 
     (*f_curpage) << "\"></div>";
 }
+#endif //disable css drawing
 
 
 } // namespace pdf2htmlEX

src/HTMLRenderer/general.cc

@@ -11,6 +11,7 @@
 #include <cmath>
 #include <algorithm>
 #include <vector>
+#include <functional>
 
 #include <GlobalParams.h>
 
@@ -46,6 +47,7 @@ HTMLRenderer::HTMLRenderer(const Param & param)
     ,html_text_page(param, all_manager)
     ,preprocessor(param)
     ,tmp_files(param)
+    ,tracer(param)
 {
     if(!(param.debug))
     {
@@ -76,6 +78,13 @@ HTMLRenderer::HTMLRenderer(const Param & param)
     all_manager.height      .set_eps(EPS);
     all_manager.width       .set_eps(EPS);
     all_manager.bottom      .set_eps(EPS);
+
+    tracer.on_char_drawn =
+            [this](double * box) { covered_text_detecor.add_char_bbox(box); };
+    tracer.on_char_clipped =
+            [this](double * box, bool partial) { covered_text_detecor.add_char_bbox_clipped(box, partial); };
+    tracer.on_non_char_drawn =
+            [this](double * box) { covered_text_detecor.add_non_char_bbox(box); };
 }
 
 HTMLRenderer::~HTMLRenderer()
@@ -133,13 +142,6 @@ void HTMLRenderer::process(PDFDoc *doc)
             cur_page_filename = filled_template_filename;
         }
 
-        if(param.process_nontext)
-        {
-            fallback_bg_required = !bg_renderer->render_page(doc, i);
-            if (fallback_bg_required && fallback_bg_renderer != nullptr)
-                fallback_bg_renderer->render_page(doc, i);
-        }
-
         doc->displayPage(this, i,
                 text_zoom_factor() * DEFAULT_DPI, text_zoom_factor() * DEFAULT_DPI,
                 0,
@@ -190,15 +192,20 @@ void HTMLRenderer::startPage(int pageNum, GfxState *state)
 void HTMLRenderer::startPage(int pageNum, GfxState *state, XRef * xref)
 #endif
 {
+    covered_text_detecor.reset();
+    tracer.reset(state);
+
     this->pageNum = pageNum;
 
-    double pageWidth = state->getPageWidth();
-    double pageHeight = state->getPageHeight();
+    html_text_page.set_page_size(state->getPageWidth(), state->getPageHeight());
 
-    html_text_page.set_page_size(pageWidth, pageHeight);
+    reset_state();
+}
+
+void HTMLRenderer::endPage() {
+    long long wid = all_manager.width.install(html_text_page.get_width());
+    long long hid = all_manager.height.install(html_text_page.get_height());
 
-    long long wid = all_manager.width.install(pageWidth);
-    long long hid = all_manager.height.install(pageHeight);
     (*f_curpage)
         << "<div id=\"" << CSS::PAGE_FRAME_CN << pageNum
             << "\" class=\"" << CSS::PAGE_FRAME_CN
@@ -231,16 +238,15 @@ void HTMLRenderer::startPage(int pageNum, GfxState *state, XRef * xref)
 
     if(param.process_nontext)
     {
-        if (!fallback_bg_required)
+        if (bg_renderer->render_page(cur_doc, pageNum))
             bg_renderer->embed_image(pageNum);
         else if (fallback_bg_renderer != nullptr)
-            fallback_bg_renderer->embed_image(pageNum);
+        {
+            if (fallback_bg_renderer->render_page(cur_doc, pageNum))
+                fallback_bg_renderer->embed_image(pageNum);
+        }
     }
 
-    reset_state();
-}
-
-void HTMLRenderer::endPage() {
     // dump all text
     html_text_page.dump_text(*f_curpage);
     html_text_page.dump_css(f_css.fs);

src/HTMLRenderer/image.cc

@@ -14,6 +14,8 @@ namespace pdf2htmlEX {
 
 void HTMLRenderer::drawImage(GfxState * state, Object * ref, Stream * str, int width, int height, GfxImageColorMap * colorMap, GBool interpolate, int *maskColors, GBool inlineImg)
 {
+    tracer.draw_image(state);
+
     return OutputDev::drawImage(state,ref,str,width,height,colorMap,interpolate,maskColors,inlineImg);
 
 #if 0
@@ -62,4 +64,20 @@ void HTMLRenderer::drawImage(GfxState * state, Object * ref, Stream * str, int w
 #endif
 }
 
+void HTMLRenderer::drawSoftMaskedImage(GfxState *state, Object *ref, Stream *str,
+                   int width, int height,
+                   GfxImageColorMap *colorMap,
+                   GBool interpolate,
+                   Stream *maskStr,
+                   int maskWidth, int maskHeight,
+                   GfxImageColorMap *maskColorMap,
+                   GBool maskInterpolate)
+{
+    tracer.draw_image(state);
+
+    return OutputDev::drawSoftMaskedImage(state,ref,str, // TODO really required?
+            width,height,colorMap,interpolate,
+            maskStr, maskWidth, maskHeight, maskColorMap, maskInterpolate);
+}
+
 } // namespace pdf2htmlEX

src/HTMLRenderer/state.cc

@@ -46,6 +46,7 @@ void HTMLRenderer::updateFont(GfxState * state)
 void HTMLRenderer::updateCTM(GfxState * state, double m11, double m12, double m21, double m22, double m31, double m32) 
 {
     ctm_changed = true; 
+    tracer.update_ctm(state, m11, m12, m21, m22, m31, m32);
 }
 void HTMLRenderer::updateTextMat(GfxState * state) 
 {
@@ -89,14 +90,17 @@ void HTMLRenderer::updateStrokeColor(GfxState * state)
 void HTMLRenderer::clip(GfxState * state)
 {
     clip_changed = true;
+    tracer.clip(state);
 }
 void HTMLRenderer::eoClip(GfxState * state)
 {
     clip_changed = true;
+    tracer.clip(state, true);
 }
 void HTMLRenderer::clipToStrokePath(GfxState * state)
 {
     clip_changed = true;
+    tracer.clip_to_stroke_path(state);
 }
 void HTMLRenderer::reset_state()
 {
@@ -119,6 +123,8 @@ void HTMLRenderer::reset_state()
     cur_line_state.y = 0;
     memcpy(cur_line_state.transform_matrix, ID_MATRIX, sizeof(cur_line_state.transform_matrix));
 
+    cur_line_state.is_char_covered = [this](int index) { return is_char_covered(index);};
+
     cur_clip_state.xmin = 0;
     cur_clip_state.xmax = 0;
     cur_clip_state.ymin = 0;
@@ -502,6 +508,10 @@ void HTMLRenderer::prepare_text_line(GfxState * state)
         double rise_x, rise_y;
         state->textTransformDelta(0, state->getRise(), &rise_x, &rise_y);
         state->transform(state->getCurX() + rise_x, state->getCurY() + rise_y, &cur_line_state.x, &cur_line_state.y);
+
+        if (param.correct_text_visibility)
+            cur_line_state.first_char_index = get_char_count();
+
         html_text_page.open_new_line(cur_line_state);
 
         cur_text_state.vertical_align = 0;

src/HTMLRenderer/text.cc

@@ -14,6 +14,9 @@
 #include "util/namespace.h"
 #include "util/unicode.h"
 
+//#define HR_DEBUG(x)  (x)
+#define HR_DEBUG(x)
+
 namespace pdf2htmlEX {
 
 using std::all_of;
@@ -51,26 +54,35 @@ void HTMLRenderer::drawString(GfxState * state, GooString * s)
     char *p = s->getCString();
     int len = s->getLength();
 
+    //accumulated displacement of chars in this string, in text object space
     double dx = 0;
     double dy = 0;
-    double dx1,dy1;
+    //displacement of current char, in text object space, including letter space but not word space.
+    double ddx, ddy;
+    //advance of current char, in glyph space
+    double ax, ay;
+    //origin of current char, in glyph space
     double ox, oy;
 
-    int nChars = 0;
-    int nSpaces = 0;
     int uLen;
 
     CharCode code;
     Unicode *u = nullptr;
 
+    HR_DEBUG(printf("HTMLRenderer::drawString:len=%d\n", len));
+
     while (len > 0) 
     {
-        auto n = font->getNextChar(p, len, &code, &u, &uLen, &dx1, &dy1, &ox, &oy);
+        auto n = font->getNextChar(p, len, &code, &u, &uLen, &ax, &ay, &ox, &oy);
+        HR_DEBUG(printf("HTMLRenderer::drawString:unicode=%lc(%d)\n", (wchar_t)u[0], u[0]));
 
         if(!(equal(ox, 0) && equal(oy, 0)))
         {
             cerr << "TODO: non-zero origins" << endl;
         }
+        ddx = ax * cur_font_size + cur_letter_space;
+        ddy = ay * cur_font_size;
+        tracer.draw_char(state, dx, dy, ax, ay);
 
         bool is_space = false;
         if (n == 1 && *p == ' ') 
@@ -85,19 +97,19 @@ void HTMLRenderer::drawString(GfxState * state, GooString * s)
              * There are always ugly PDF files with no useful info at all.
              */
             is_space = true;
-            ++nSpaces;
         }
         
         if(is_space && (param.space_as_offset))
         {
+            html_text_page.get_cur_line()->append_padding_char();
             // ignore horiz_scaling, as it has been merged into CTM
-            html_text_page.get_cur_line()->append_offset((dx1 * cur_font_size + cur_letter_space + cur_word_space) * draw_text_scale); 
+            html_text_page.get_cur_line()->append_offset((ax * cur_font_size + cur_letter_space + cur_word_space) * draw_text_scale);
         }
         else
         {
             if((param.decompose_ligature) && (uLen > 1) && all_of(u, u+uLen, isLegalUnicode))
             {
-                html_text_page.get_cur_line()->append_unicodes(u, uLen, (dx1 * cur_font_size + cur_letter_space));
+                html_text_page.get_cur_line()->append_unicodes(u, uLen, ddx);
             }
             else
             {
@@ -110,7 +122,7 @@ void HTMLRenderer::drawString(GfxState * state, GooString * s)
                 {
                     uu = unicode_from_font(code, font);
                 }
-                html_text_page.get_cur_line()->append_unicodes(&uu, 1, (dx1 * cur_font_size + cur_letter_space));
+                html_text_page.get_cur_line()->append_unicodes(&uu, 1, ddx);
                 /*
                  * In PDF, word_space is appended if (n == 1 and *p = ' ')
                  * but in HTML, word_space is appended if (uu == ' ')
@@ -123,19 +135,15 @@ void HTMLRenderer::drawString(GfxState * state, GooString * s)
             }
         }
 
-        dx += dx1;
-        dy += dy1;
+        dx += ddx * cur_horiz_scaling;
+        dy += ddy;
+        if (is_space)
+            dx += cur_word_space * cur_horiz_scaling;
 
-        ++nChars;
         p += n;
         len -= n;
     }
 
-    // horiz_scaling is merged into ctm now, 
-    // so the coordinate system is ugly
-    dx = (dx * cur_font_size + nChars * cur_letter_space + nSpaces * cur_word_space) * cur_horiz_scaling;
-    dy *= cur_font_size;
-
     cur_tx += dx;
     cur_ty += dy;
         
@@ -143,4 +151,16 @@ void HTMLRenderer::drawString(GfxState * state, GooString * s)
     draw_ty += dy;
 }
 
+bool HTMLRenderer::is_char_covered(int index)
+{
+    auto covered = covered_text_detecor.get_chars_covered();
+    if (index < 0 || index >= (int)covered.size())
+    {
+        std::cerr << "Warning: HTMLRenderer::is_char_covered: index out of bound: "
+                << index << ", size: " << covered.size() <<endl;
+        return false;
+    }
+    return covered[index];
+}
+
 } // namespace pdf2htmlEX

src/HTMLState.h

@@ -5,6 +5,8 @@
 #ifndef HTMLSTATE_H__
 #define HTMLSTATE_H__
 
+#include <functional>
+
 #include "Color.h"
 
 namespace pdf2htmlEX {
@@ -62,6 +64,12 @@ struct HTMLLineState
 {
     double x,y;
     double transform_matrix[4];
+    // The page-cope char index(in drawing order) of the first char in this line.
+    int first_char_index;
+    // A function to determine whether a char is covered at a given index.
+    std::function<bool(int)> is_char_covered;
+
+    HTMLLineState(): first_char_index(-1) { }
 };
 
 struct HTMLClipState

src/HTMLTextLine.cc

@@ -36,7 +36,14 @@ HTMLTextLine::HTMLTextLine (const HTMLLineState & line_state, const Param & para
 
 void HTMLTextLine::append_unicodes(const Unicode * u, int l, double width)
 {
-    text.insert(text.end(), u, u+l);
+    if (l == 1)
+        text.push_back(min(u[0], (unsigned)INT_MAX));
+    else if (l > 1)
+    {
+        text.push_back(- decomposed_text.size() - 1);
+        decomposed_text.emplace_back();
+        decomposed_text.back().assign(u, u + l);
+    }
     this->width += width;
 }
 
@@ -69,6 +76,60 @@ void HTMLTextLine::append_state(const HTMLTextState & text_state)
     last_state.font_size *= last_state.font_info->font_size_scale;
 }
 
+void HTMLTextLine::dump_char(std::ostream & out, int pos)
+{
+    int c = text[pos];
+    if (c > 0)
+    {
+        Unicode u = c;
+        writeUnicodes(out, &u, 1);
+    }
+    else if (c < 0)
+    {
+        auto dt = decomposed_text[- c - 1];
+        writeUnicodes(out, &dt.front(), dt.size());
+    }
+}
+
+void HTMLTextLine::dump_chars(ostream & out, int begin, int len)
+{
+    static const Color transparent(0, 0, 0, true);
+
+    if (line_state.first_char_index < 0)
+    {
+        for (int i = 0; i < len; i++)
+            dump_char(out, begin + i);
+        return;
+    }
+
+    bool invisible_group_open = false;
+    for(int i = 0; i < len; i++)
+    {
+        if (!line_state.is_char_covered(line_state.first_char_index + begin + i)) //visible
+        {
+            if (invisible_group_open)
+            {
+                invisible_group_open = false;
+                out << "</span>";
+            }
+            dump_char(out, begin + i);
+        }
+        else
+        {
+            if (!invisible_group_open)
+            {
+                out << "<span class=\"" << all_manager.fill_color.get_css_class_name()
+                    << all_manager.fill_color.install(transparent) << " " << all_manager.stroke_color.get_css_class_name()
+                    << all_manager.stroke_color.install(transparent) << "\">";
+                invisible_group_open = true;
+            }
+            dump_char(out, begin + i);
+        }
+    }
+    if (invisible_group_open)
+        out << "</span>";
+}
+
 void HTMLTextLine::dump_text(ostream & out)
 {
     /*
@@ -216,7 +277,7 @@ void HTMLTextLine::dump_text(ostream & out)
                 size_t next_text_idx = text_idx2;
                 if((cur_offset_iter != offsets.end()) && (cur_offset_iter->start_idx) < next_text_idx)
                     next_text_idx = cur_offset_iter->start_idx;
-                writeUnicodes(out, (&text.front()) + cur_text_idx, next_text_idx - cur_text_idx);
+                dump_chars(out, cur_text_idx, next_text_idx - cur_text_idx);
                 cur_text_idx = next_text_idx;
             }
         }

src/HTMLTextLine.h

@@ -73,7 +73,16 @@ public:
         double width;
     };
 
+    /**
+     * Append a drawn char (glyph)'s unicode. l > 1 mean this glyph correspond to
+     * multiple code points.
+     */
     void append_unicodes(const Unicode * u, int l, double width);
+    /**
+     * Append a special padding char with 0 width, in order to keep char index consistent.
+     * The padding char is ignored during output.
+     */
+    void append_padding_char() { text.push_back(0); }
     void append_offset(double width);
     void append_state(const HTMLTextState & text_state);
     void dump_text(std::ostream & out);
@@ -92,6 +101,13 @@ private:
     void optimize_normal(std::vector<HTMLTextLine*> &);
     void optimize_aggressive(std::vector<HTMLTextLine*> &);
 
+    /**
+     * Dump chars' unicode to output stream.
+     * begin/pos is the index in 'text'.
+     */
+    void dump_chars(std::ostream & out, int begin, int len);
+    void dump_char(std::ostream & out, int pos);
+
     const Param & param;
     AllStateManager & all_manager;
 
@@ -102,7 +118,16 @@ private:
 
     std::vector<State> states;
     std::vector<Offset> offsets;
-    std::vector<Unicode> text;
+
+    /**
+     * Drawn chars (glyph) in this line are stored in 'text'. For each element c in 'text':
+     * - If c > 0, it is the unicode code point corresponds to the glyph;
+     * - If c == 0, it is a padding char, and ignored during output (TODO some bad PDFs utilize 0?);
+     * - If c < -1, this glyph corresponds to more than one unicode code points,
+     *   which are stored in 'decomposed_text', and (-c-1) is the index in 'decomposed_text'.
+     */
+    std::vector<int> text;
+    std::vector<std::vector<Unicode> > decomposed_text;
 };
 
 } // namespace pdf2htmlEX

src/HTMLTextPage.h

@@ -39,6 +39,9 @@ public:
     void set_page_size(double width, double height);
     void clip(const HTMLClipState & clip_state);
 
+    double get_width() { return page_width; }
+    double get_height() { return page_height; }
+
 private:
     void optimize(void);
 

src/Param.h

@@ -38,6 +38,7 @@ struct Param
     int process_nontext;
     int process_outline;
     int process_annotation;
+    int correct_text_visibility;
     int printing;
     int fallback;
     int tmp_file_size_limit;

src/pdf2htmlEX.cc

@@ -187,6 +187,7 @@ void parse_options (int argc, char **argv)
         .add("space-as-offset", &param.space_as_offset, 0, "treat space characters as offsets")
         .add("tounicode", &param.tounicode, 0, "how to handle ToUnicode CMaps (0=auto, 1=force, -1=ignore)")
         .add("optimize-text", &param.optimize_text, 0, "try to reduce the number of HTML elements used for text")
+        .add("correct-text-visibility", &param.correct_text_visibility, 0, "try to detect texts covered by other graphics and properly arrange them")
 
         // background image
         .add("bg-format", &param.bg_format, "png", "specify background image format")

src/util/math.cc

@@ -1,8 +1,12 @@
 #include <cstring>
 #include <limits>
+#include <algorithm>
 
 #include "math.h"
 
+using std::min;
+using std::max;
+
 namespace pdf2htmlEX {
 
 void tm_transform(const double * tm, double & x, double & y, bool is_delta)
@@ -56,5 +60,31 @@ void tm_transform_bbox(const double * tm, double * bbox)
     }
 }
 
+bool bbox_intersect(const double * bbox1, const double * bbox2, double * result)
+{
+    double x0, y0, x1, y1;
+
+    x0 = max(min(bbox1[0], bbox1[2]), min(bbox2[0], bbox2[2]));
+    x1 = min(max(bbox1[0], bbox1[2]), max(bbox2[0], bbox2[2]));
+
+    if (x0 >= x1)
+        return false;
+
+    y0 = max(min(bbox1[1], bbox1[3]), min(bbox2[1], bbox2[3]));
+    y1 = min(max(bbox1[1], bbox1[3]), max(bbox2[1], bbox2[3]));
+
+    if (y0 >= y1)
+        return false;
+
+    if (result)
+    {
+        result[0] = x0;
+        result[1] = y0;
+        result[2] = x1;
+        result[3] = y1;
+    }
+    return true;
+}
+
 } //namespace pdf2htmlEX 
 

src/util/math.h

@@ -24,6 +24,13 @@ static inline bool tm_equal(const double * tm1, const double * tm2, int size = 6
             return false;
     return true;
 }
+
+static inline void tm_init(double * tm)
+{
+    tm[0] = tm[3] = 1;
+    tm[1] = tm[2] = tm[4] = tm[5] = 0;
+}
+
 static inline void tm_multiply(double * result, const double * m1, const double * m2)
 {
     result[0] = m1[0] * m2[0] + m1[2] * m2[1];
@@ -39,6 +46,14 @@ static inline double hypot(double x, double y) { return std::sqrt(x*x+y*y); }
 void tm_transform(const double * tm, double & x, double & y, bool is_delta = false);
 void tm_multiply(double * tm_left, const double * tm_right);
 void tm_transform_bbox(const double * tm, double * bbox);
+/**
+ * Calculate the intersection of 2 boxes.
+ * If they are intersecting, store the result to result (if not null) and return true.
+ * Otherwise return false, and result is not touched.
+ * Param result can be same as one of bbox1 and bbox2.
+ * Data in boxes are expected in the order of (x0, y0, x1, y1).
+ */
+bool bbox_intersect(const double * bbox1, const double * bbox2, double * result = nullptr);
 
 } //namespace pdf2htmlEX 
 #endif //MATH_H__