Leadtools OCR
This project benchmarks and compares the performance and accuracy of two leading OCR (Optical Character Recognition) engines: Aspose.OCR and Leadtools.OCR. The goal is to evaluate their effectiveness on a real-world dataset and provide clear, quantitative results for both speed and recognition quality.
This benchmark uses:
- Aspose.OCR for .NET 25.7.0 (release date: 14.07.2025)
- Leadtools.Ocr 23.0.0.4 (release date: 02.01.2025)
Licensing Requirements
Both Aspose.OCR and Leadtools.OCR are commercial products and require valid licenses to operate in full-feature mode.
- Aspose.OCR: Requires a valid license file (e.g.,
Aspose.OCR.NET2025_2.lic
). - Leadtools.OCR: Requires a valid license and key file (e.g.,
LEADTOOLS.lic
andLEADTOOLS.lic.key
).
Licenses must be properly installed and referenced in the application for the engines to function without limitations or watermarks.
Dataset
- The benchmark was conducted on a dataset containing 994 images .
- Each image has a corresponding ground-truth text file for accuracy evaluation.
Benchmark Results
Engine | Average Time per Image | Levenshtein Error (%) |
---|---|---|
Aspose | 2.25 seconds | 3.27% |
Aspose (using AI Spell-check) | 62.3 seconds | 3.15% |
Leadtools | 0.99 seconds | 5.13% |
- Time per Image: Average time taken to process a single image.
- Levenshtein Error (%): Average percentage difference between the recognized text and the ground-truth, measured using the Levenshtein distance (lower is better).
Interpretation
- Aspose.OCR provides slightly better recognition accuracy (lower Levenshtein error), but is slower.
- Leadtools.OCR is significantly faster, with a small trade-off in recognition accuracy.
Example 1
Original Image 3717 symbols![]() |
Aspose OCR vs Aspose OCR AI
OCR 0.43% errors | OCR + LLM 0.27% errors |
---|---|
![]() | ![]() |
Leadtools OCR
0.7% errors |
---|
![]() |
Key Differences and Analysis
Type of Difference | Aspose | Aspose (AI) | Leadtools | Ethalon |
---|---|---|---|---|
Punctuation | ll7(c) | 117(c) | ll7(c), 60 days , | 117(c) |
Case Sensitivity | C&c, DoE, c&c | C&c | the DOE, C8 |
C&C, DOE |
Character Substitution | none significant | none significant | VolLl.l'lll€ I, C86, C8 |
n/a |
Word Substitution | none | none | C86 agreements | n/a |
Missing Content | none | none | none | complete |
Example 2
Original Image 968 symbols![]() |
Aspose OCR vs Aspose OCR AI
OCR 3.51% errors | OCR + LLM 2.79% errors |
---|---|
![]() | ![]() |
Leadtools OCR
65.5% errors |
---|
![]() |
Key Differences and Analysis
Type of Difference | Aspose | Aspose (AI) | Leadtools | Ethalon |
---|---|---|---|---|
Punctuation | August 19. 1993 | August 19. 1993 | lots of garbled punctuation | August 19, 1993 |
Case Sensitivity | sincerely, AsMc | sincerely, AsMc | Random casing and symbols | Sincerely, ABMC, Illinois |
Character Substitution | Il, Medica1, We'17 | Il | ALE>q/\N, MP. Rica 4 | IL, Medical, We'll |
Word Substitution | Fer_more, cati us | None | Heavy distortion | For more, call us |
Missing Content | None | None | Meaning lost due to garbling | None |
- Aspose: Very close to the ethalon, with only minor punctuation and case errors. No missing content.
- Leadtools: Slightly less accurate, with a typo (“t1ons”), a word substitution (“systan operation”), and more case inconsistencies.
Aspose is more reliable for this sample, as reflected in the lower Levenshtein error (0.83% vs 1.7%).
Easy to use Aspose.OCR API
Aspose OCR API simplifies text extraction by handling image analysis and processing automatically, letting users focus on their core tasks—just provide an image and and leave the rest to the API.
Convert image to text
/** Initialize recognition API */
AsposeOcr api = new AsposeOcr();
/** Provide the image for recognition */
OcrInput source = new OcrInput(InputType.SingleImage);
source.Add("image.png");
/** Recognize and output results */
List<RecognitionResult> results = api.Recognize(source);
Console.WriteLine(results[0].RecognitionText);