Forensic Source Attribution of Color Laser Printers from Machine Identification Code Patterns: Impact of Class Imbalance on Machine Learning Performance
International Journal of Medical Toxicology and Forensic Medicine,
Vol. 16 (2026),
1 January 2026
,
Page 1-10
https://doi.org/10.22037/ijmtfm.v16.53598
Abstract
Background: Printer source attribution cases are among the most critically benchmarked on artificially balanced datasets, a practice that masks the classifiers. This further complicates the study when one manufacturer dominates the evidence pool. This research study further frames the source identification as a class-imbalanced problem and uses features like geometric and textural structures of the Machine Identification Code (MIC) dot patterns from the samples collected across six major color laser printer manufacturers under the skewed, uncurated class distribution of the printed outputs.
Methods: Nine interpretable descriptors spanning geometric complexity (Fractal Dimension, Laplacian Variance, and Standard Deviation) and Gray-Level Co-occurrence Matrix texture (Contrast, Correlation, Homogeneity, Energy, Entropy) were extracted and evaluated first through unsupervised clustering and then through five supervised classifiers. The unsupervised clustering failed to recover brand identity from the raw feature space, confirming that the proximity- based grouping alone cannot resolve the manufacturer overlap.
Results: Among the supervised models, Random Forest achieved the strongest balanced performance and the highest accuracy, consistently outperforming the distance-based and single-tree baselines. Furthermore, per-class analysis exposed a pronounced imbalance vulnerability, while the majority class and texturally distinctive brands, including Konica, Ricoh, and HP were identified with near-perfect recall. The minority- class brand Toshiba was misclassified in every model (recall = 0), and Xerox recall remained below 33% throughout.
Conclusion: These results indicate that the reported aggregate accuracy substantially overstates the real- world reliability for the underrepresented printer sources, and they motivate the imbalance training strategies as a necessary next step for the forensic-grade printer attribution systems rather than an optional refinement.
- Class Imbalance
- Ensemble Learning
- Document Authentication
- Gray Level Co-occurrence Matrix
- Texture Analysis
- Printer Forensics
- Machine Identification Code
How to Cite
References
[1] Choi JH, Lee HY, Lee HK. Color laser printer forensics based on noisy features and a support vector machine classifier. Multimed Tools Appl. 2013;67(2):363-82. [DOI: 10.1007/s11042-011-0835-9]
[2] Zhao Y, Yang X, Chen X. A review of laser printer classification and identification. Forensic Sci Int. 2025;371:112480. [DOI: 10.1016/j.forsciint.2025.112480]
[3] Escher S, Strafe T. Robustness analysis of a passive printer identification scheme for halftone images. In: Proceedings of the IEEE International Conference on Image Processing (ICIP); 2017 Sep 17-20; Beijing, China. IEEE; 2017. p. 4357-61. [DOI: 10.1109/ICIP.2017.8297059]
[4] Mikkilineni AK, Chiang PJ, Ali GN, Chiu GTC, Allebach JP, Delp EJ III. Printer identification based on gray-level co-occurrence features for security and forensic applications. Proc SPIE. 2005;5681:430-40. [DOI: 10.1117/12.593796]
[5] Orhan B, Mao J, Sharma G. 2009 IEEE International Conference on Acoustics, Speech and Signal Processing: proceedings; 2009 Apr 19-24; Taipei, Taiwan. IEEE; 2009. [DOI: 10.1109/ICASSP.2009.5099042]
[6] Khanna N, Mikkilineni AK, Chiu GTC, Allebach JP, Delp EJ. Survey of scanner and printer forensics at Purdue University. Lect Notes Comput Sci. 2008;5158:22-34. [DOI: 10.1007/978-3-540-85303-9_3]
[7] Tsai MJ, Yuadi I, Tao YH. Decision-theoretic model to identify printed sources. Multimed Tools Appl. 2018;77(21):27543-87. [DOI: 10.1007/s11042-018-5938-0]
[8] Salim AS, Abdalla AA. Application of Adobe Photoshop CC 2018 for identifying the source of HP color laser printouts. Arab J Forensic Sci Forensic Med. 2019;1(2):1158-64. [DOI: 10.26735/16586794.2019.002]
[9] Salim AS, Abdalla AA. Determination of the identity and uniqueness of Ricoh-brand color laser printouts using Adobe Creative Cloud Photoshop 2018. Egypt J Forensic Sci. 2019;9(1):16. [DOI: 10.1186/s41935-019-0140-8]
[10] Salim AS, Abdalla AA. The state-of-the-art technique for determining the identity and uniqueness of common color laser printouts by coded dot matrix patterns. Arab J Forensic Sci Forensic Med. 2023;5(1):58-73. [DOI: 10.26735/RUGH9293]
[11] Ferreira A, Bondi L, Baroffio L, Bestagini P, Huang J, Dos Santos JA, et al. Data-driven feature characterization techniques for laser printer attribution. IEEE Trans Inf Forensics Secur. 2017;12(8):1860-73. [DOI: 10.1109/TIFS.2017.2692722]
[12] Choi JH, Lee HK, Lee HY, Suh YH. Color laser printer forensics with noise texture analysis. In: Proceedings of the 2010 ACM SIGMM Multimedia Security Workshop; 2010 Oct 26; Firenze, Italy. ACM; 2010. p. 19-24. [DOI: 10.1145/1854229.1854235]
[13] Shang S, Kong X. Printer and scanner forensics. In: Handbook of digital forensics of multimedia data and devices. Chichester: John Wiley & Sons; 2015. p. 375-410. [DOI: 10.1002/9781118705773.ch10]
[14] Chiang PJ, Ali GN, Mikkilineni AK, Delp EJ, Allebach JP, Chiu GTC. Extrinsic signatures embedding and detection for information hiding and secure printing in electrophotography. In: Proceedings of the American Control Conference; 2006 June 14-16; Minneapolis, MN, USA. IEEE; 2006. p. 2539-44. [DOI: 10.1109/ACC.2006.1656557]
[15] Hamzehyan R, Razzazi F, Behrad A. Printer source identification by feature modeling in the total variable printer space. J Forensic Sci. 2021;66(6):2261-73. [DOI: 10.1111/1556-4029.14822]
[16] Devi MU. Statistical measures for differentiation of photocopy from print technology: forensic perspective. Int J Comput Appl. 2014;105(1):1-7. [DOI: 10.5120/18328-9366]
[17] Tsai MJ, Tao YH, Yuadi I. Deep learning for printed document source identification. Signal Process Image Commun. 2019;70:184-98. [DOI: 10.1016/j.image.2018.09.006]
[18] Liu N, Zhang L, Chen C. A preliminary study on sources of banding artifacts for the identification of monochromatic laser printers. J Forensic Sci. 2017;62(6):1542-53. [DOI: 10.1111/1556-4029.13487]
[19] Arora S, Sarin RK, Puri P. Machine identification codes of color laser printers: revisiting privacy and security. Ind J Electrical Eng Comput Sci. 2025;40(1):137-45. [DOI: 10.11591/ijeecs.v40.i1.pp137-145]
[20] Salim AS, Abdalla AAM. Application of Adobe Photoshop CC 2018 for identifying color laser printer source of Xerox brand. Egypt J Forensic Sci. 2018;8(1):52. [DOI: 10.1186/s41935-018-0076-4]
[21] Verhoeven G, Karel W, Štuhec S, Doneus M, Trinks I, Pfeifer N. Mind your grey tones: examining the influence of decolorization methods on interest point extraction and matching for architectural image-based modeling. Int Arch Photogramm Remote Sens Spat Inf Sci. 2015;XL-5-W4:307-14. [DOI: 10.5194/isprsarchives-XL-5-W4-307-2015]
[22] Yang P, Baracchi D, Ni R, Zhao Y, Argenti F, Piva A. A survey of deep learning-based source image forensics. J Imaging. 2020;6(9):94. [DOI: 10.3390/jimaging6090094]
[23] Lee HY, Choi JH. Printer identification using intrinsic noise. In: Proceedings of the 5th International Conference on Ubiquitous Information Technologies and Applications; 2010 Dec 16-18; Sanya, China. IEEE; 2010. [DOI: 10.1109/ICUT.2010.5677893]
[24] Jiang ZF, Zhang QH, Wang YC, Liu YL, Zhao YW, Hao YY, et al. Prediction of laser printers and cartridges based on three-dimensional profiles via discrimination analysis. Forensic Sci Int. 2024;363:112186. [DOI: 10.1016/j.forsciint.2024.112186]
[25] Gál L, Oravec M, Gemeiner P, Čeppan M. Principal component analysis for the forensic discrimination of black inkjet inks based on the Vis-NIR fiber optics reflection spectra. Forensic Sci Int. 2015;257:285-92. [DOI: 10.1016/j.forsciint.2015.09.011]
[26] Karna A, Gibert K. Automatic identification of the number of clusters in hierarchical clustering. Neural Comput Appl. 2022;34(1):119-34. [DOI: 10.1007/s00521-021-05873-3]
[27] Kim DG, Hou JU, Lee HK. Learning deep features for source color laser printer identification based on cascaded learning. Neurocomputing. 2019;365:219-28. [DOI: 10.1016/j.neucom.2019.07.084]
[28] Elkasrawi S, Shafait F. Printer identification using supervised learning for document forgery detection. In: Proceedings of the 11th IAPR International Workshop on Document Analysis Systems (DAS); 2014 Apr 7-10; Tours, France. IEEE; 2014. p. 146-50. [DOI: 10.1109/DAS.2014.23]
[29] El Abady NF, Zayed HH, Taha M. An efficient source printer identification model using convolution neural network (SPI-CNN). Int J Adv Comput Sci Appl. 2023;14(3):745-53. [DOI: 10.14569/IJACSA.2023.0140386]
[30] Navarro LC, Navarro AKW, Rocha A, Dahab R. Connecting the dots: toward accountable machine-learning printer attribution methods. J Vis Commun Image Represent. 2018;53:257-72. [DOI: 10.1016/j.jvcir.2018.04.002]
[31] Jamieson A, Moenssens AA, editors. Wiley encyclopedia of forensic science. Chichester: John Wiley & Sons; 2009. [DOI: 10.1002/9780470061589]
[32] Phan-Ho AT, Nguyen QT, Patrix J, Verny J. Source printer identification with microscopic printing using deep learning. IFAC-PapersOnLine. 2022;55(13):1177-82. [DOI: 10.1016/j.ifacol.2022.10.395]
[33] Darwish SM, Elgohary HM. Building an expert system for printer forensics: a new printer identification model based on niching genetic algorithm. Expert Syst. 2021;38(4):e12624. [DOI: 10.1111/exsy.12624]
- Abstract Viewed: 0 times
- 53598 Downloaded: 0 times