pdf_oxide 0.3.59

The fastest Rust PDF library with text extraction: 0.8ms mean, 100% pass rate on 3,830 PDFs. 5× faster than pdf_extract, 17× faster than oxidize_pdf. Extract, create, and edit PDFs.
Documentation
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
//! Locate Form XObject invocations on a PDF page and compute the
//! page-space bbox each one occupies.
//!
//! Existing image extraction in `pdf_to_ir::extract_page_images` only
//! covers raster `Image` XObjects (`/Subtype /Image`). Many cover
//! pages — federal regs (CFR), agency logos, dissertation banners —
//! draw small vector decorations as `/Subtype /Form` XObjects
//! instead. Without surfacing those we leave a visible hole in the
//! office round-trip.
//!
//! This module walks the page content stream for `Do` operators
//! whose name resolves to a Form XObject in the page resources.
//! Each invocation captures the current transformation matrix (CTM)
//! at the point of invocation; combined with the Form's own `/BBox`
//! (and optional `/Matrix`), it gives the page-space rectangle the
//! Form draws into.
//!
//! The caller then renders that rectangle as a bitmap (via
//! `pdf_oxide::rendering::render_page_region`) and embeds the result
//! into the IR as an `Image`. Output: every Form XObject on the page
//! becomes a positioned raster image at the source coordinates.

use crate::content::{parse_content_stream, Operator};
use crate::object::{Object, ObjectRef};
use std::collections::HashMap;

/// One Form-XObject invocation discovered on a page. Bounds are in
/// PDF user space (origin bottom-left, y-up).
#[derive(Debug, Clone)]
pub struct FormInvocation {
    /// Resource name the page used to invoke the Form
    /// (`Do` operand, e.g. `"Fm0"`).
    #[allow(dead_code)]
    pub name: String,
    /// PDF user-space bbox: (x_lower_left, y_lower_left, width, height).
    pub bbox_pt: (f32, f32, f32, f32),
}

/// Scan `page_idx`'s content stream for inline image (`BI…ID…EI`)
/// blocks. Each inline image's bbox is the unit square (0,0)-(1,1)
/// transformed through the current CTM at the BI point, per PDF
/// spec §8.9.7.
///
/// Inline images on cover pages are typically 1-bit raster icons —
/// agency logos, GPO marks, accessibility badges — that don't
/// warrant a full Image XObject. Without surfacing them the office
/// round-trip drops the icon entirely.
pub fn find_inline_image_invocations(
    doc: &crate::document::PdfDocument,
    page_idx: usize,
) -> Vec<FormInvocation> {
    let content_bytes = match doc.get_page_content_data(page_idx) {
        Ok(b) => b,
        Err(_) => return Vec::new(),
    };
    let ops = match parse_content_stream(&content_bytes) {
        Ok(v) => v,
        Err(_) => return Vec::new(),
    };

    let mut ctm: [f32; 6] = [1.0, 0.0, 0.0, 1.0, 0.0, 0.0];
    let mut stack: Vec<[f32; 6]> = Vec::new();
    let mut out = Vec::new();
    let mut inline_idx = 0usize;
    for op in ops {
        match op {
            Operator::SaveState => {
                stack.push(ctm);
            },
            Operator::RestoreState => {
                if let Some(prev) = stack.pop() {
                    ctm = prev;
                }
            },
            Operator::Cm { a, b, c, d, e, f } => {
                ctm = matrix_multiply(&[a, b, c, d, e, f], &ctm);
            },
            Operator::InlineImage { .. } => {
                // Inline image source space is the unit square.
                let bbox_pt = transform_bbox(&(0.0, 0.0, 1.0, 1.0), &ctm);
                out.push(FormInvocation {
                    name: format!("__inline_{inline_idx}"),
                    bbox_pt,
                });
                inline_idx += 1;
            },
            _ => {},
        }
    }
    out
}

/// Scan `page_idx`'s content stream for `Do` operators that invoke
/// Form XObjects in the page's `/Resources/XObject` dictionary.
/// Returns the page-space bbox each Form occupies.
pub fn find_form_xobject_invocations(
    doc: &crate::document::PdfDocument,
    page_idx: usize,
) -> Vec<FormInvocation> {
    // Fetch page resources/XObject dict so we can filter Do operands
    // to only Form-type XObjects (skip raster Images already handled
    // by `extract_page_images`).
    let xobject_dict = match get_page_xobjects(doc, page_idx) {
        Some(d) => d,
        None => return Vec::new(),
    };

    // Map name → Form-XObject info (BBox, optional Matrix). Names
    // not present here are either raster Images (skipped here) or
    // unresolvable refs.
    let mut form_info: HashMap<String, FormGeometry> = HashMap::new();
    for (name, value) in &xobject_dict {
        let obj_ref = match value.as_reference() {
            Some(r) => r,
            None => continue,
        };
        if let Some(geom) = load_form_geometry(doc, obj_ref) {
            form_info.insert(name.clone(), geom);
        }
    }
    if form_info.is_empty() {
        return Vec::new();
    }

    // Pull the page's content stream bytes. `get_page_content_data`
    // joins all stream parts in order and decompresses them.
    let content_bytes = match doc.get_page_content_data(page_idx) {
        Ok(b) => b,
        Err(_) => return Vec::new(),
    };
    let ops = match parse_content_stream(&content_bytes) {
        Ok(v) => v,
        Err(_) => return Vec::new(),
    };

    // Walk the operator stream tracking the CTM. PDF spec §8.3.4:
    //   - `q` saves graphics state; `Q` restores.
    //   - `cm` concatenates a 6-element matrix to the current CTM.
    //   - `Do` invokes the named XObject, transforming its own
    //     coordinate system through the current CTM (and the form's
    //     own `/Matrix` if present) before drawing.
    let mut ctm: [f32; 6] = [1.0, 0.0, 0.0, 1.0, 0.0, 0.0];
    let mut stack: Vec<[f32; 6]> = Vec::new();
    let mut out = Vec::new();
    for op in ops {
        match op {
            Operator::SaveState => {
                stack.push(ctm);
            },
            Operator::RestoreState => {
                if let Some(prev) = stack.pop() {
                    ctm = prev;
                }
            },
            Operator::Cm { a, b, c, d, e, f } => {
                ctm = matrix_multiply(&[a, b, c, d, e, f], &ctm);
            },
            Operator::Do { name } => {
                if let Some(geom) = form_info.get(&name) {
                    let mut effective = ctm;
                    if let Some(form_matrix) = geom.matrix {
                        effective = matrix_multiply(&form_matrix, &ctm);
                    }
                    let bbox_pt = transform_bbox(&geom.bbox, &effective);
                    out.push(FormInvocation { name, bbox_pt });
                }
            },
            _ => {},
        }
    }
    out
}

struct FormGeometry {
    /// Form-space BBox: (llx, lly, urx, ury).
    bbox: (f32, f32, f32, f32),
    /// Optional /Matrix mapping form space → user space.
    matrix: Option<[f32; 6]>,
}

fn load_form_geometry(
    doc: &crate::document::PdfDocument,
    obj_ref: ObjectRef,
) -> Option<FormGeometry> {
    let obj = doc.load_object(obj_ref).ok()?;
    let dict = obj.as_dict()?;
    if dict.get("Subtype").and_then(|s| s.as_name()) != Some("Form") {
        return None;
    }
    let bbox_arr = dict.get("BBox").and_then(|b| b.as_array())?;
    if bbox_arr.len() != 4 {
        return None;
    }
    let bbox = (
        as_number(&bbox_arr[0])?,
        as_number(&bbox_arr[1])?,
        as_number(&bbox_arr[2])?,
        as_number(&bbox_arr[3])?,
    );
    let matrix = dict
        .get("Matrix")
        .and_then(|m| m.as_array())
        .and_then(|arr| {
            if arr.len() != 6 {
                return None;
            }
            Some([
                as_number(&arr[0])?,
                as_number(&arr[1])?,
                as_number(&arr[2])?,
                as_number(&arr[3])?,
                as_number(&arr[4])?,
                as_number(&arr[5])?,
            ])
        });
    Some(FormGeometry { bbox, matrix })
}

fn get_page_xobjects(
    doc: &crate::document::PdfDocument,
    page_idx: usize,
) -> Option<HashMap<String, Object>> {
    let page = doc.get_page(page_idx).ok()?;
    let page_dict = page.as_dict()?;
    let resources = match page_dict.get("Resources") {
        Some(r) => match r.as_reference() {
            Some(rref) => doc.load_object(rref).ok()?,
            None => r.clone(),
        },
        None => return None,
    };
    let res_dict = resources.as_dict()?;
    let xobjects = match res_dict.get("XObject") {
        Some(x) => match x.as_reference() {
            Some(xref) => doc.load_object(xref).ok()?,
            None => x.clone(),
        },
        None => return None,
    };
    let dict = xobjects.as_dict()?;
    Some(dict.clone())
}

fn as_number(obj: &Object) -> Option<f32> {
    match obj {
        Object::Integer(i) => Some(*i as f32),
        Object::Real(f) => Some(*f as f32),
        _ => None,
    }
}

/// PDF matrices: row-major in spec but stored column-major in
/// content streams as [a b c d e f] meaning
///   | a b 0 |
///   | c d 0 |
///   | e f 1 |
/// (concatenating: `cm` postmultiplies onto the current CTM).
fn matrix_multiply(a: &[f32; 6], b: &[f32; 6]) -> [f32; 6] {
    [
        a[0] * b[0] + a[1] * b[2],
        a[0] * b[1] + a[1] * b[3],
        a[2] * b[0] + a[3] * b[2],
        a[2] * b[1] + a[3] * b[3],
        a[4] * b[0] + a[5] * b[2] + b[4],
        a[4] * b[1] + a[5] * b[3] + b[5],
    ]
}

/// Rasterise every Form-XObject + inline-image region on `page_idx`
/// at 200 DPI as PNG bytes, returning each one paired with its
/// page-space bbox in PDF user space (origin bottom-left, y-up).
///
/// Used by both the flow-mode path (`pdf_to_ir::extract_page_images`)
/// and the layout-mode writers (`docx_layout`, `pptx_layout`,
/// `xlsx_layout`). Without it, vector figures embedded as Form
/// XObjects — common in academic papers and government docs — are
/// dropped from the office round-trip entirely.
///
/// `existing_rects_pdf` carries the bboxes of raster Image XObjects
/// already extracted by the caller (PDF y-up coords); any
/// Form/inline region that overlaps one of those by more than 50% of
/// its own area is skipped to avoid double-rasterising (e.g. the
/// NARA seal referenced both as Image and via a Form alias).
#[cfg(feature = "rendering")]
pub fn rasterize_form_and_inline_regions(
    doc: &crate::document::PdfDocument,
    page_idx: usize,
    page_h_pt: f32,
    existing_rects_pdf: &[(f32, f32, f32, f32)],
) -> Vec<((f32, f32, f32, f32), Vec<u8>)> {
    use crate::rendering::{render_page, ImageFormat as RFmt, RenderOptions};

    let mut invs = find_form_xobject_invocations(doc, page_idx);
    invs.extend(find_inline_image_invocations(doc, page_idx));

    // Same heuristics as `pdf_to_ir::extract_page_images`:
    // - drop pixel-tiny boxes (probably parser artefacts);
    // - drop boxes wider than 1.5× page height or taller than the
    //   page itself (extreme outliers, likely CTM noise);
    // - drop regions that significantly overlap an already-extracted
    //   raster Image (avoid rendering the seal twice).
    invs.retain(|inv| {
        let (_, _, w, h) = inv.bbox_pt;
        w >= 4.0 && h >= 4.0 && w < page_h_pt * 1.5 && h < page_h_pt
    });
    let overlaps = |bbox: &(f32, f32, f32, f32)| -> bool {
        let (ix, iy, iw, ih) = *bbox;
        existing_rects_pdf.iter().any(|(rx, ry, rw, rh)| {
            let l = ix.max(*rx);
            let r = (ix + iw).min(rx + rw);
            let b = iy.max(*ry);
            let t = (iy + ih).min(ry + rh);
            let inter = (r - l).max(0.0) * (t - b).max(0.0);
            let area = iw * ih;
            area > 0.0 && inter / area > 0.5
        })
    };
    invs.retain(|inv| !overlaps(&inv.bbox_pt));
    if invs.is_empty() {
        return Vec::new();
    }

    // Render the full page ONCE, then crop in image space for each
    // region. Previously we called `render_page_region` per region;
    // that helper does a full-page render + crop internally, so a
    // page with N Form-XObjects re-rendered the page N times. On a
    // 10-page LaTeX paper with 12 Form-XObjects we saw 263 seconds
    // for PDF→DOCX; with this caching it should drop by ~Nx.
    let bytes = doc.source_bytes.clone();
    if bytes.is_empty() {
        return Vec::new();
    }
    let doc_mut = match crate::document::PdfDocument::from_bytes(bytes) {
        Ok(d) => d,
        Err(_) => return Vec::new(),
    };
    let dpi: u32 = 150;
    let opts = RenderOptions {
        dpi,
        format: RFmt::Png,
        ..Default::default()
    };
    let full = match render_page(&doc_mut, page_idx, &opts) {
        Ok(i) => i,
        Err(_) => return Vec::new(),
    };
    let full_img = match image::load_from_memory(&full.data) {
        Ok(i) => i,
        Err(_) => return Vec::new(),
    };
    let scale = dpi as f32 / 72.0;
    let img_w = full_img.width();
    let img_h = full_img.height();

    let mut out = Vec::with_capacity(invs.len());
    for inv in invs {
        let (x_pdf, y_pdf, w, h) = inv.bbox_pt;
        // PDF y-up → image y-down. Image origin is top-left.
        let top_y_pt = page_h_pt - (y_pdf + h);
        let cx = (x_pdf * scale).round().max(0.0) as u32;
        let cy = (top_y_pt * scale).round().max(0.0) as u32;
        let cw = (w * scale).round().max(1.0) as u32;
        let ch = (h * scale).round().max(1.0) as u32;
        let x = cx.min(img_w.saturating_sub(1));
        let y = cy.min(img_h.saturating_sub(1));
        let cw = cw.min(img_w - x);
        let ch = ch.min(img_h - y);
        if cw == 0 || ch == 0 {
            continue;
        }
        let cropped = full_img.crop_imm(x, y, cw, ch);
        let mut buf = Vec::new();
        use image::codecs::png::{CompressionType, FilterType, PngEncoder};
        use image::ImageEncoder;
        if PngEncoder::new_with_quality(&mut buf, CompressionType::Fast, FilterType::Sub)
            .write_image(cropped.as_bytes(), cw, ch, cropped.color().into())
            .is_err()
        {
            continue;
        }
        if buf.is_empty() {
            continue;
        }
        out.push(((x_pdf, y_pdf, w, h), buf));
    }
    out
}

/// Transform a form-space BBox (llx, lly, urx, ury) through the
/// effective matrix → page-space (x, y, width, height). Computes
/// all four corners and returns the axis-aligned bounding rectangle.
fn transform_bbox(bbox: &(f32, f32, f32, f32), m: &[f32; 6]) -> (f32, f32, f32, f32) {
    let corners = [
        (bbox.0, bbox.1),
        (bbox.2, bbox.1),
        (bbox.2, bbox.3),
        (bbox.0, bbox.3),
    ];
    let mut min_x = f32::MAX;
    let mut min_y = f32::MAX;
    let mut max_x = f32::MIN;
    let mut max_y = f32::MIN;
    for (x, y) in corners {
        let tx = m[0] * x + m[2] * y + m[4];
        let ty = m[1] * x + m[3] * y + m[5];
        if tx < min_x {
            min_x = tx;
        }
        if tx > max_x {
            max_x = tx;
        }
        if ty < min_y {
            min_y = ty;
        }
        if ty > max_y {
            max_y = ty;
        }
    }
    (min_x, min_y, (max_x - min_x).max(0.0), (max_y - min_y).max(0.0))
}