Skip to content

A fast, helpful, and open-source document parser

undo CLI arg changes, make image filtering more broad, make types make sense

Logan Markewich710a604ec617
diff --git a/cli/parse.ts b/cli/parse.ts
index 38a1a55..b5ee4c6 100644
--- a/cli/parse.ts
+++ b/cli/parse.ts
@@ -26,7 +26,6 @@ interface ParseCommandOptions {
dpi?: string;
preciseBbox?: boolean;
preserveSmallText?: boolean;
- detectImages?: boolean;
password?: string;
config?: string;
quiet?: boolean;
@@ -82,7 +81,6 @@ program
.option("--dpi <dpi>", "DPI for rendering", DEFAULT_DPI.toString())
.option("--no-precise-bbox", "Disable precise bounding boxes")
.option("--preserve-small-text", "Preserve very small text")
- .option("--detect-images", "Detect embedded images (logos, photos, charts) via PDFium")
.option("--password <password>", "Password for encrypted/protected documents")
.option("--config <file>", "Config file (JSON)")
.option("-q, --quiet", "Suppress progress output")
@@ -127,7 +125,6 @@ program
dpi: parseInt(options.dpi || DEFAULT_DPI.toString()),
preciseBoundingBox: options.preciseBbox !== false,
preserveVerySmallText: options.preserveSmallText || false,
- detectImages: options.detectImages || false,
password: options.password,
};
diff --git a/src/core/config.ts b/src/core/config.ts
index 902c140..a9bc5e8 100644
--- a/src/core/config.ts
+++ b/src/core/config.ts
@@ -21,7 +21,6 @@ export const DEFAULT_CONFIG: LiteParseConfig = {
preserveVerySmallText: false,
preserveLayoutAlignmentAcrossPages: false,
password: undefined,
- detectImages: false,
};
export function mergeConfig(userConfig: Partial<LiteParseConfig>): LiteParseConfig {
diff --git a/src/core/parser.ts b/src/core/parser.ts
index 3911c47..ddb35af 100644
--- a/src/core/parser.ts
+++ b/src/core/parser.ts
@@ -70,7 +70,7 @@ export class LiteParse {
this.config = mergeConfig(userConfig);
// Initialize PDF engine
- this.pdfEngine = new PdfJsEngine({ detectImages: this.config.detectImages });
+ this.pdfEngine = new PdfJsEngine();
// Initialize OCR engine
// Auto-detect: use HTTP OCR if URL provided, otherwise use Tesseract
diff --git a/src/core/types.ts b/src/core/types.ts
index d1d9995..a32593b 100644
--- a/src/core/types.ts
+++ b/src/core/types.ts
@@ -136,15 +136,6 @@ export interface LiteParseConfig {
* @defaultValue `undefined`
*/
password?: string;
-
- /**
- * Detect embedded images (logos, photos, charts) using PDFium and populate
- * the `images` array on each page. When enabled, this also allows OCR to
- * trigger on pages that contain images but have sufficient native text.
- *
- * @defaultValue `false`
- */
- detectImages: boolean;
}
/**
diff --git a/src/engines/pdf/pdfium-renderer.ts b/src/engines/pdf/pdfium-renderer.ts
index 2b4f293..80b17b9 100644
--- a/src/engines/pdf/pdfium-renderer.ts
+++ b/src/engines/pdf/pdfium-renderer.ts
@@ -2,6 +2,37 @@ import { PDFiumLibrary, type PDFiumPageRenderOptions } from "@hyzyla/pdfium";
import sharp from "sharp";
import { promises as fs } from "fs";
+/**
+ * Minimal interface for PDFium's WASM module internals.
+ * These properties are private/protected in @hyzyla/pdfium's typings,
+ * but we need direct access for low-level page object bound queries.
+ */
+interface PdfiumWasmModule {
+ _FPDFPageObj_GetBounds?: (
+ objHandle: number,
+ leftPtr: number,
+ bottomPtr: number,
+ rightPtr: number,
+ topPtr: number
+ ) => number;
+ _FPDF_GetPageWidthF: (pagePtr: number) => number;
+ _FPDF_GetPageHeightF: (pagePtr: number) => number;
+ _malloc: (size: number) => number;
+ _free: (ptr: number) => void;
+ HEAPU8: Uint8Array;
+}
+
+/** Minimum image dimension in PDF points to be considered for OCR */
+const MIN_IMAGE_SIZE_PT = 25;
+/** Images covering more than this fraction of the page are treated as backgrounds */
+const MAX_IMAGE_PAGE_COVERAGE = 0.9;
+
+interface PdfiumPageInternal {
+ module: PdfiumWasmModule;
+ pageIdx: number;
+ objects(): Iterable<{ type: string; objectIdx: number }>;
+}
+
/**
* PDFium-based PDF screenshot renderer
* Uses native PDFium library for high-quality, fast screenshots
@@ -91,30 +122,27 @@ export class PdfiumRenderer {
const document = await this.pdfium.loadDocument(pdfBuffer, password);
try {
- const page = document.getPage(pageNumber - 1);
+ // Cast to internal interface — module/pageIdx/objectIdx are private in
+ // @hyzyla/pdfium typings but required for low-level WASM bound queries
+ const page = document.getPage(pageNumber - 1) as unknown as PdfiumPageInternal;
const results: Array<{ x: number; y: number; width: number; height: number }> = [];
- // Access the underlying WASM module for low-level bound queries
- // eslint-disable-next-line @typescript-eslint/no-explicit-any
- const mod = (page as any).module;
- // eslint-disable-next-line @typescript-eslint/no-explicit-any
- const pagePtr = (page as any).pageIdx;
+ const mod = page.module;
+ const pagePtr = page.pageIdx;
if (!mod || !mod._FPDFPageObj_GetBounds) {
return results;
}
// Get page dimensions via WASM (getSize() is private in typings)
- const pageWidth: number = mod._FPDF_GetPageWidthF(pagePtr);
- const pageHeight: number = mod._FPDF_GetPageHeightF(pagePtr);
+ const pageWidth = mod._FPDF_GetPageWidthF(pagePtr);
+ const pageHeight = mod._FPDF_GetPageHeightF(pagePtr);
// Iterate page objects looking for images
for (const obj of page.objects()) {
if (obj.type !== "image") continue;
- // objectIdx is the WASM object handle/pointer directly
- // eslint-disable-next-line @typescript-eslint/no-explicit-any
- const objHandle = (obj as any).objectIdx;
+ const objHandle = obj.objectIdx;
if (!objHandle) continue;
// Allocate memory for 4 floats (left, bottom, right, top)
@@ -133,9 +161,9 @@ export class PdfiumRenderer {
const w = right - left;
const h = top - bottom;
- // Filter out tiny images (< 10pt) and full-page backgrounds (> 90% coverage)
- if (w < 10 || h < 10) continue;
- if (w > pageWidth * 0.9 && h > pageHeight * 0.9) continue;
+ if (w < MIN_IMAGE_SIZE_PT || h < MIN_IMAGE_SIZE_PT) continue;
+ if (w > pageWidth * MAX_IMAGE_PAGE_COVERAGE && h > pageHeight * MAX_IMAGE_PAGE_COVERAGE)
+ continue;
// Convert PDF coords (Y-up) to viewport coords (Y-down)
results.push({
diff --git a/src/engines/pdf/pdfjs.ts b/src/engines/pdf/pdfjs.ts
index c233e43..06001c2 100644
--- a/src/engines/pdf/pdfjs.ts
+++ b/src/engines/pdf/pdfjs.ts
@@ -432,11 +432,6 @@ export class PdfJsEngine implements PdfEngine {
private pdfiumRenderer: PdfiumRenderer | null = null;
private currentPdfPath: string | null = null;
private currentPdfData: Uint8Array | null = null;
- private detectImages: boolean;
-
- constructor(options?: { detectImages?: boolean }) {
- this.detectImages = options?.detectImages ?? false;
- }
async loadDocument(input: string | Uint8Array, password?: string): Promise<PdfDocument> {
let data: Uint8Array;
@@ -605,22 +600,20 @@ export class PdfJsEngine implements PdfEngine {
}
let images: Image[] = [];
- if (this.detectImages) {
- try {
- if (!this.pdfiumRenderer) {
- this.pdfiumRenderer = new PdfiumRenderer();
- }
- const pdfInput = this.currentPdfPath || this.currentPdfData || doc.data;
- const imageBounds = await this.pdfiumRenderer.extractImageBounds(pdfInput, pageNum);
- images = imageBounds.map((bounds) => ({
- x: bounds.x,
- y: bounds.y,
- width: bounds.width,
- height: bounds.height,
- }));
- } catch {
- // Image extraction is best-effort
+ try {
+ if (!this.pdfiumRenderer) {
+ this.pdfiumRenderer = new PdfiumRenderer();
}
+ const pdfInput = this.currentPdfPath || this.currentPdfData || doc.data;
+ const imageBounds = await this.pdfiumRenderer.extractImageBounds(pdfInput, pageNum);
+ images = imageBounds.map((bounds) => ({
+ x: bounds.x,
+ y: bounds.y,
+ width: bounds.width,
+ height: bounds.height,
+ }));
+ } catch {
+ // Image extraction is best-effort
}
// Skip annotation extraction - not currently used in processing pipeline