/*! * pixaJoin() * * Input: pixad (dest pixa; add to this one) * pixas (source pixa; add from this one) * istart (starting index in nas) * iend (ending index in nas; use 0 to cat all) * Return: 0 if OK, 1 on error * * Notes: * (1) This appends a clone of each indicated pix in pixas to pixad * (2) istart < 0 is taken to mean 'read from the start' (istart = 0) * (3) iend <= 0 means 'read to the end' */ l_int32 pixaJoin(PIXA *pixad, PIXA *pixas, l_int32 istart, l_int32 iend) { l_int32 ns, i; BOXA *boxas, *boxad; PIX *pix; PROCNAME("pixaJoin"); if (!pixad) return ERROR_INT("pixad not defined", procName, 1); if (!pixas) return ERROR_INT("pixas not defined", procName, 1); if ((ns = pixaGetCount(pixas)) == 0) { L_INFO("empty pixas", procName); return 0; } if (istart < 0) istart = 0; if (istart >= ns) return ERROR_INT("istart out of bounds", procName, 1); if (iend <= 0) iend = ns - 1; if (iend >= ns) return ERROR_INT("iend out of bounds", procName, 1); if (istart > iend) return ERROR_INT("istart > iend; nothing to add", procName, 1); for (i = istart; i <= iend; i++) { pix = pixaGetPix(pixas, i, L_CLONE); pixaAddPix(pixad, pix, L_INSERT); } boxas = pixaGetBoxa(pixas, L_CLONE); boxad = pixaGetBoxa(pixad, L_CLONE); boxaJoin(boxad, boxas, 0, 0); boxaDestroy(&boxas); /* just the clones */ boxaDestroy(&boxad); return 0; }
std::vector<Figure> extractFigures(PIX *original, PageRegions &pageRegions, DocumentStatistics &docStats, bool verbose, bool showSteps, std::vector<Figure> &errors) { BOXA *bodytext = pageRegions.bodytext; BOXA *graphics = pageRegions.graphics; BOXA *captions = pageRegions.getCaptionsBoxa(); std::vector<Caption> unassigned_captions = pageRegions.captions; int total_captions = captions->n; PIXA *steps = showSteps ? pixaCreate(4) : NULL; // Add bodyText boxes to fill up the margin BOX *margin; BOX *foreground; pixClipToForeground(original, NULL, &foreground); BOX *extent; boxaGetExtent(graphics, NULL, NULL, &extent); margin = boxBoundingRegion(extent, foreground); boxDestroy(&extent); boxaGetExtent(bodytext, NULL, NULL, &extent); margin = boxBoundingRegion(margin, extent); boxDestroy(&extent); boxaGetExtent(pageRegions.other, NULL, NULL, &extent); margin = boxBoundingRegion(margin, extent); int x = margin->x - 2, y = margin->y - 2, h = margin->h + 4, w = margin->w + 4; x = std::max(x, 0); y = std::max(y, 0); h = std::min((int)original->h, h); w = std::min((int)original->w, w); boxDestroy(&margin); boxaAddBox(bodytext, boxCreate(0, 0, original->w, y), L_CLONE); boxaAddBox(bodytext, boxCreate(0, y + h, original->w, original->h - y - h), L_CLONE); boxaAddBox(bodytext, boxCreate(0, 0, x, original->h), L_CLONE); boxaAddBox(bodytext, boxCreate(x + w, 0, original->w - x - w, original->h), L_CLONE); // Add captions to body text boxaJoin(bodytext, captions, 0, captions->n); if (showSteps) pixaAddPix(steps, original, L_CLONE); // Generate proposed regions for each caption box double center = original->w / 2.0; BOXAA *allProposals = boxaaCreate(captions->n); BOXA *claimedImages = boxaCreate(captions->n); for (int i = 0; i < captions->n; i++) { BOX *captBox = boxaGetBox(captions, i, L_CLONE); BOXA *proposals = boxaCreate(4); for (int j = 0; j < bodytext->n; j++) { BOX *txtBox = boxaGetBox(bodytext, j, L_CLONE); BOX *proposal = NULL; int tolerance = 2; int horizontal = 0; int vertical = 0; boxAlignment(captBox, txtBox, tolerance, &horizontal, &vertical); if (vertical * horizontal != 0 or (vertical == 0 and horizontal == 0)) { continue; } if (vertical == 0) { if (horizontal == 1) { proposal = boxRelocateOneSide(NULL, captBox, txtBox->x + txtBox->w + 2, L_FROM_LEFT); } else if (horizontal == -1) { proposal = boxRelocateOneSide(NULL, captBox, txtBox->x - 2, L_FROM_RIGHT); } boxExpandUD(proposal, bodytext); if (horizontal == -1) { proposal->w -= captBox->w + 1; proposal->x = captBox->x + captBox->w + 1; } else if (horizontal == 1) { proposal->w -= captBox->w + 1; } } else { if (vertical == 1) { proposal = boxRelocateOneSide(NULL, captBox, txtBox->y + txtBox->h + 3, L_FROM_TOP); } else if (vertical == -1) { proposal = boxRelocateOneSide(NULL, captBox, txtBox->y - 3, L_FROM_BOT); } boxExpandLR(proposal, bodytext); if (vertical == -1) { proposal->h -= captBox->h + 1; proposal->y = captBox->y + captBox->h + 1; } else if (vertical == 1) { proposal->h -= captBox->h + 1; } } // For two columns document, captions that do not // cross the center should not have regions pass the center if (docStats.documentIsTwoColumn()) { if (captBox->x + captBox->w <= center and proposal->x + proposal->w > center) { boxRelocateOneSide(proposal, proposal, center - 1, L_FROM_RIGHT); } else if (captBox->x >= center and proposal->x < center) { boxRelocateOneSide(proposal, proposal, center + 1, L_FROM_LEFT); } } BOX *clippedProposal; pixClipBoxToForeground(original, proposal, NULL, &clippedProposal); if (clippedProposal != NULL and scoreBox(clippedProposal, pageRegions.captions.at(i).type, bodytext, graphics, claimedImages, original) > 0) { boxaAddBox(proposals, clippedProposal, L_CLONE); } } if (proposals->n > 0) { boxaaAddBoxa(allProposals, proposals, L_CLONE); } else { // Give up on this caption int on_caption = i - (total_captions - unassigned_captions.size()); errors.push_back(Figure(unassigned_captions.at(on_caption), NULL)); unassigned_captions.erase(unassigned_captions.begin() + on_caption); } } std::vector<Figure> figures = std::vector<Figure>(); if (unassigned_captions.size() == 0) { return figures; } // Now go through every possible assignment of captions // to proposals pick the highest scorign one int numConfigurations = 1; for (int i = 0; i < allProposals->n; ++i) { numConfigurations *= allProposals->boxa[i]->n; } if (verbose) printf("Found %d possible configurations\n", numConfigurations); BOXA *bestProposals = NULL; std::vector<bool> bestKeep; int bestFound = -1; double bestScore = -1; for (int onConfig = 0; onConfig < numConfigurations; ++onConfig) { // Gather the proposed regions based on the configuration number int configNum = onConfig; BOXA *proposals = boxaCreate(allProposals->n); std::vector<bool> keep; for (int i = 0; i < allProposals->n; ++i) { int numProposals = allProposals->boxa[i]->n; int selected = configNum % numProposals; configNum = configNum / numProposals; boxaAddBox(proposals, allProposals->boxa[i]->box[selected], L_COPY); } // Attempt to split any overlapping regions for (int i = 0; i < proposals->n; ++i) { for (int j = i; j < proposals->n; ++j) { BOX *p1 = proposals->box[i]; BOX *p2 = proposals->box[j]; int eq; boxEqual(p1, p2, &eq); if (not eq) continue; int vertical, horizontal; boxAlignment(unassigned_captions.at(i).boundingBox, unassigned_captions.at(j).boundingBox, 2, &horizontal, &vertical); if (vertical == 0 or horizontal != 0) continue; double split = splitBoxVertical(original, p1); if (split > 0) { BOX *topClipped; BOX *botClipped; BOX *top = boxRelocateOneSide(NULL, p1, split - 1, L_FROM_BOT); pixClipBoxToForeground(original, top, NULL, &topClipped); BOX *bot = boxRelocateOneSide(NULL, p1, split + 1, L_FROM_TOP); pixClipBoxToForeground(original, bot, NULL, &botClipped); if (vertical == -1) { proposals->box[i] = topClipped; proposals->box[j] = botClipped; } else { proposals->box[i] = botClipped; proposals->box[j] = topClipped; } if (verbose) printf("Split a region vertically\n"); } } } if (showSteps) { pixaAddPix(steps, pixDrawBoxa(original, proposals, 4, 0xff000000), L_CLONE); } // Score the proposals int numFound = 0; double totalScore = 0; for (int i = 0; i < proposals->n; ++i) { double score = scoreBox(proposals->box[i], pageRegions.captions.at(i).type, bodytext, graphics, proposals, original); totalScore += score; if (score > 0) { numFound += 1; keep.push_back(true); } else { keep.push_back(false); } } // Switch in for the current best needed if (numFound > bestFound or (numFound == bestFound and totalScore > bestScore)) { bestFound = numFound; bestScore = totalScore; bestProposals = proposals; bestKeep = keep; } } if (showSteps) { BOX *clip; PIXA *show = pixaCreate(4); pixClipBoxToForeground(original, NULL, NULL, &clip); int pad = 10; clip->x -= 10; clip->y -= 10; clip->w += pad * 2; clip->h += pad * 2; for (int i = 0; i < steps->n; ++i) { pixaAddPix(show, pixClipRectangle(steps->pix[i], clip, NULL), L_CLONE); } pixDisplay(pixaDisplayTiled(pixaConvertTo32(show), 4000, 1, 30), 0, 0); } for (int i = 0; i < bestProposals->n; ++i) { if (bestKeep.at(i)) { BOX *imageBox = bestProposals->box[i]; int pad = 2; imageBox->x -= pad; imageBox->y -= pad; imageBox->w += pad * 2; imageBox->h += pad * 2; figures.push_back(Figure(unassigned_captions.at(i), imageBox)); } else { errors.push_back(Figure(unassigned_captions.at(i), NULL)); } } return figures; }
/*! * pixSplitIntoCharacters() * * Input: pixs (1 bpp, contains only deskewed text) * minw (minimum component width for initial filtering; typ. 4) * minh (minimum component height for initial filtering; typ. 4) * &boxa (<optional return> character bounding boxes) * &pixa (<optional return> character images) * &pixdebug (<optional return> showing splittings) * * Return: 0 if OK, 1 on error * * Notes: * (1) This is a simple function that attempts to find split points * based on vertical pixel profiles. * (2) It should be given an image that has an arbitrary number * of text characters. * (3) The returned pixa includes the boxes from which the * (possibly split) components are extracted. */ l_int32 pixSplitIntoCharacters(PIX *pixs, l_int32 minw, l_int32 minh, BOXA **pboxa, PIXA **ppixa, PIX **ppixdebug) { l_int32 ncomp, i, xoff, yoff; BOXA *boxa1, *boxa2, *boxat1, *boxat2, *boxad; BOXAA *baa; PIX *pix, *pix1, *pix2, *pixdb; PIXA *pixa1, *pixadb; PROCNAME("pixSplitIntoCharacters"); if (pboxa) *pboxa = NULL; if (ppixa) *ppixa = NULL; if (ppixdebug) *ppixdebug = NULL; if (!pixs || pixGetDepth(pixs) != 1) return ERROR_INT("pixs not defined or not 1 bpp", procName, 1); /* Remove the small stuff */ pix1 = pixSelectBySize(pixs, minw, minh, 8, L_SELECT_IF_BOTH, L_SELECT_IF_GT, NULL); /* Small vertical close for consolidation */ pix2 = pixMorphSequence(pix1, "c1.10", 0); pixDestroy(&pix1); /* Get the 8-connected components */ boxa1 = pixConnComp(pix2, &pixa1, 8); pixDestroy(&pix2); boxaDestroy(&boxa1); /* Split the components if obvious */ ncomp = pixaGetCount(pixa1); boxa2 = boxaCreate(ncomp); pixadb = (ppixdebug) ? pixaCreate(ncomp) : NULL; for (i = 0; i < ncomp; i++) { pix = pixaGetPix(pixa1, i, L_CLONE); if (ppixdebug) { boxat1 = pixSplitComponentWithProfile(pix, 10, 7, &pixdb); if (pixdb) pixaAddPix(pixadb, pixdb, L_INSERT); } else { boxat1 = pixSplitComponentWithProfile(pix, 10, 7, NULL); } pixaGetBoxGeometry(pixa1, i, &xoff, &yoff, NULL, NULL); boxat2 = boxaTransform(boxat1, xoff, yoff, 1.0, 1.0); boxaJoin(boxa2, boxat2, 0, -1); pixDestroy(&pix); boxaDestroy(&boxat1); boxaDestroy(&boxat2); } pixaDestroy(&pixa1); /* Generate the debug image */ if (ppixdebug) { if (pixaGetCount(pixadb) > 0) { *ppixdebug = pixaDisplayTiledInRows(pixadb, 32, 1500, 1.0, 0, 20, 1); } pixaDestroy(&pixadb); } /* Do a 2D sort on the bounding boxes, and flatten the result to 1D */ baa = boxaSort2d(boxa2, NULL, 0, 0, 5); boxad = boxaaFlattenToBoxa(baa, NULL, L_CLONE); boxaaDestroy(&baa); boxaDestroy(&boxa2); /* Optionally extract the pieces from the input image */ if (ppixa) *ppixa = pixClipRectangles(pixs, boxad); if (pboxa) *pboxa = boxad; else boxaDestroy(&boxad); return 0; }