Volume 20. Number 1
December 2021

Original Articles

StarGAN Deepfake Videos Detection Based on No-reference Image Quality Assessments and Support Vector Machine

Wen-Chao Yang

Abstract¡GWith the rapid development of digital technology, Deepfake can be used in multimedia fields, such as face replacement, image forgery, and synthesized speech. In particular, the widespread use of generative adversarial networks (GANs) makes it difficult for current image detection and multimedia forensics technologies to identify authenticity. The improper application of Deepfake technology will lead to severe consequences.In this paper, a new Deepfake detection method based on no-reference image quality assessment (NR-IQA) features and a support vector machine (SVM) auto-classifier is proposed. The proposed method is used to detect fake facial images generated using the StarGAN model. The experimental results show that the proposed method has the best results of 7-fold cross-validation based on SVM with the linear kernel using two PIQE features, and its detection error rate is 0.15.

Keywords: forensic science; artificial intelligence; deepfake video; StarGAN; multimedia forensics; no-reference image quality assessment 


Using Next-Generation Sequencing to Analyze Mixed Forensic Specimens

Yung-Chun Lai ; Man-Chen Chang.; Kuan-Cheng Peng; Chun-Yen Lin

Abstract¡G Capillary electrophoresis (CE) has routinely been used in forensic DNA laboratories to analyze STRs and DNA sequences. However, this technique has limited ability to discriminate mixed forensic specimens. Most such specimens have very little DNA, and their mixing ratio is usually unknown. For this reason, it remains difficult to identify trace amounts of DNA in mixed samples using CE. Next-generation sequencing (NGS) can overcome the disparity problem of the DNA mixing ratio in samples by increasing the sequencing depth. In this study, 15 forensic specimens were analyzed for STR and Y-STR DNA profiles using NGS, and the results were then compared with CE data to evaluate its correctness and reliability. First, 15 samples were collected from forensic autopsy cases and were analyzed via CE. The remaining DNA samples were analyzed using NGS technology. Each sample had a different amount of DNA in the NGS analysis, depending on the condition of the forensic case. The detection rates of the STR DNA allele of NGS in 13 cases were all greater than CE, with the exception of two samples (AS1 and NA1) with low amounts of DNA (<1 ng) for NGS analysis. Furthermore, among the 15 specimens, four specimens (NA1, NA2, VS1, and VS10) were also examined for human mitochondrial HV1 and HV2 sequences. Using NGS to analyze the mitochondrial DNA of mixed forensic samples can evade the challenge of quantifying the DNA mixing ratio using CE. This study suggests that the depth of NGS sequencing must be increased, supplemented by the proportion of SNP or mitochondrial bases, to effectively solve the mixed-profile problem.

 Keywords: forensic science; next-generation sequencing (NGS); capillary electrophoresis (CE); mixed forensic specimens; human identification; mitochondrial DNA


Profiling Fingerprint Patterns in Singapore across Race and Gender for Forensic Applications

Kwok-Pui Choi; Wei-Ling, Stella Tan; Shi-Yun Tan; Jessica Gunawan; Huilin Rochelle NG

Abstract¡G Previous empirical studies conducted have suggested a possible correlation between fingerprint patterns and gender and between fingerprint patterns and race. However, no such studies have been carried out in Singapore. Thus, this study aims to profile the fingerprint patterns in Singapore across different genders and races, to determine the relationship between fingerprint patterns and gender as well as between fingerprint patterns and race. Based on this study of ours, the proportion of fingerprint patterns on all thumbs and fingers are independent of an individual¡¦s gender. On the other hand, our results suggest that the proportion of fingerprint patterns on the left thumb, left middle finger, and left ring finger depend on one¡¦s race. It was also observed that the most common fingerprint pattern on the thumbs (right: 56.0%, left: 50.0%), index fingers (right: 48.2%, left: 47.5%), and ring fingers (right: 60.0%, left: 62.4%) are whorls, whereas the most common fingerprint pattern on our middle (right: 63.1%, left: 58.9%) and little fingers (right: 71.6%, left: 76.2%) are ulnar loops. Looking only at gender, most female and male subjects have whorls on all thumbs and fingers. However, taking only race into account, most Chinese subjects have whorls on all thumbs and fingers. In the case of Indian and Malay subjects, most have ulnar loops on all thumbs and fingers.

Keywords: forensic science; fingerprint; fingerprint pattern profiles; race; gender; identification; human hand bilateral symmetry


Overview of Methods for Vaginal Fluid Identification

Chia-Tzu Hsu; Li-Chin Tsai; Kuo-Lan Liu; Nu-En Huang; Yu-Hsuan Chang; Hsing-Mei Hsieh

Abstract¡GBiological evidence can connect suspects and victims to crime scenes and provide clues for reconstructing events. For example, vaginal fluid is a crucial indicator of sexual assault, especially in cases where semen stains are not found. Identifying the vaginal fluids on the fingers of perpetrators or on the objects used to violate victims can help validate victims¡¦ claims. However, unlike for semen, there are still no standard protocols for vaginal fluid identification. Forensic scientists have developed various methods to distinguish vaginal fluids from other body fluids, such as detecting vaginal cells, female hormones, specific proteins, microbes, and nucleic acids in the sample. This article provides an overview of the methods, from traditional to molecular, that has been studied in this regard and concludes all the identified targets and markers that are useful.

Keywords: forensic science; forensic biology; body fluid identification; vaginal fluids; sexual assault