pdfimgextract 1.6.4

Fast parallel PDF image extractor

python-pdfimgextract

🚀 Overview

python-pdfimgextract is a high-performance PDF image extraction tool designed to maximize throughput through multiprocessing.

The project focuses on speed, reliability, and efficient parallel processing, enabling large PDFs with hundreds of high-resolution images to be processed significantly faster than traditional single-threaded approaches.

---

✨ Features

Feature	Description
⚡ Parallel Extraction	Utilizes multiprocessing to decode images across multiple CPU cores
🛡️ Atomic Writes	Prevents partially written files during crashes
🧹 Deduplication	Optional removal of identical images
📊 Progress Tracking	Real-time progress bar during extraction
🛑 Graceful Interrupts	Safe handling of SIGINT and termination signals
💻 Clean CLI	Simple command-line interface

---

📥 Installation

pip install pdfimgextract

---

🛠️ Usage

Basic usage:

pdfimgextract INPUT_PDF OUTPUT_DIR NUMBER_OF_PROCESSES

Example:

pdfimgextract manga.pdf output 16

Optional usage (flags):

pdfimgextract -i input.pdf -o output_dir -p 8
pdfimgextract -i input.pdf -o output_dir -d hash
pdfimgextract -i input.pdf -o output_dir --overwrite

Arguments

-i / --input         Path to input PDF
-o / --output        Output directory
-p / --parallelism   Number of worker processes (default: 8)
-d / --dedup         Deduplication method: xref (default) or hash (precise but slower)
--overwrite          Overwrite existing files

If not specified, the tool defaults to:

• 8 workers
• xref deduplication

---

📊 Performance & Benchmark

🖥️ Test Environment

• OS: Windows 11
• CPU: 28 cores
• Input File: 491 MB PDF
• Extracted Images: 230
• Image Size Range: ~2MB – 10MB

---

📈 Results

Proc	Time (s)	Speedup	Efficiency	RAM (MB)
1	111.88	1.00x	100%	349
2	56.98	1.96x	98%	626
4	32.41	3.45x	86%	1159
8	20.16	5.55x	69%	2254
16	14.24	7.86x	49%	4423
32	11.41	9.81x	31%	7309
64	11.67	9.59x	15%	9306

---

🧠 Analysis

Scaling

• Near-linear scaling up to ~4 processes
• Strong gains up to ~16
• Performance plateaus around ~32

CPU Efficiency

• Very high at low parallelism (~98% at 2 workers)
• Gradual drop due to scheduling and contention
• Diminishing returns at 64 and beyond workers

Memory Usage

• RAM scales almost linearly with process count
• Each worker holds its own decoding state
• Large images amplify memory consumption

Examples:

• 8 workers → ~2.2 GB
• 32 workers → ~7.3 GB
• 64 workers → ~9.3 GB

I/O Bottleneck

• Parallel writes saturate disk bandwidth
• Causes worker stalls at high process counts
• Main limiter beyond ~32 workers

---

🏁 Optimal Range

Recommended configuration:

8 – 16 workers

Best balance between:

• Speed
• Efficiency
• Memory usage
• I/O pressure