Binarization of Image of Firearm Land Impressions Using a Convolutional Neural Network

The aim of this work is to develop an algorithm for automatically converting grayscale rifling marks into binary data in the way they are analyzed by an expert. The objective of this study is to explore the feasibility of image binarization of secondary traces using a semantic binary segmentation algorithm implemented on the basis of a convolutional neural network. The training set included 160 images of secondary traces, represented in grayscale and formed by different rifling marks. The secondary trace image generator was used to augment the training set. The calculations done in accordance with the metrics showed the following accuracy of image binarization of secondary traces: Accuracy = 0,88 ± 0,05; Recall = 0,76 ± 0,06; Precision = 0,83 ± 0,06; F1 = 0,79 ± 0,05; MCC = 0,71 ± 0,06. The binarization of images is necessary for automatic comparison of secondary traces using one of the machine learning algorithms for the purpose of their binary classification into the categories of “matching traces” and “non-matching traces”. The preliminary examination showed that the classification accuracy of pairs of secondary trace images by these categories, carried out using the Random Forest algorithm, was around 84–86%.

1. Dil’din Yu.M., Martynov V.V. Typical Expert Methods for Examination of Physical Evidence. Part 1. Moscow : INTERKRIM-PRESS, 2010. 568 p. (In Russ.).

2. Kokin A.V., Yarmak K.V. Forensic Ballistics and Forensic Ballistic Examination. Textbook. Moscow: Yuniti-dana: Zakon i pravo, 2017. 350 p. (In Russ.).

3. Chen Z., Chu W., Soons J.A., Thompson R.M., Song J., Zhao X. Fired Bullet Signature Correlation Using the Congruent Matching Profile Segments (CMPS) Method. Forensic Science International. 2019. Vol. 305. P. 10–19. https://doi.org/10.1016/j.forsciint.2019.109964

4. Chen Z., Song J., Soons J.A., Thompson R.M., Zhao X. Pilot Study on Deformed Bullet Correlation. Forensic Science International. 2020. Vol.306. P. 1–11. https://doi.org/10.1016/j.forsciint.2019.110098

5. Biasotti A.A. Statistical Study of the Individual Characteristics of Fired Bullets. Journal of Forensic Sciences. 1959. Vol. 4. No. 1. P. 34–50.

6. Fedorenko V.A., Navrotskaya E.V. Criteria and Algorithm of the Evaluation of the Uniqueness of the Complexes of Matching Tracks in the Traces on the Shot Bullets. Journal of Information Technologies and Computation Systems. 2019. No. 1. P. 110–120. (In Russ.).

7. Hare E., Hofmann H., Carriquiry A. Algorithmic Approaches to Match Degraded Land Impressions. Law, Probability and Risk. 2017. Vol. 16. No. 4. P. 203–221. https://doi.org/10.1093/lpr/mgx018

8. Hare E., Hofmann H., Carriquiry A. Automatic Matching of Bullet Land Impressions. The Annals of Applied Statistics. 2017. Vol. 11. No. 4. P. 2332–2356. https://doi.org/10.1214/17-AOAS1080

9. Fedorenko V., Sorokina K., Giverts P. Analysis of Traces on Discharged Bullets by the Congruent Matching Profile Segments Method and k-Nearest Neighbors. Programming and Com-puter Software. 2023. Vol. 49. No. 2. P. 72–81. https://doi.org/10.1134/S036176882310002X

10. Fedorenko V.A., Sorokina K.O., Giverts P.V. Analysis of Traces on the Discharged Bullets by the Congruent Matching Profile Segments Method and k-Nearest Neighbors. Journal of Information Technologies and Computation Systems. 2021. No. 1. P. 70–82. (In Russ.). https://doi.org/10.14357/20718632210108

11. Gonsales R., Vuds R. The World of Digital Processing. Digital Image Processing. P.A. Chochia (ed.). Moscow: Tekhnosfera, 2019. 1104 p. (In Russ.).

12. Kitov V.V. Data Augmentation. Machine and Deep Learning. Online Textbook. (In Russ.). https://deepmachinelearning.ru/docs/Neural-networks/Regularization/Augmentation