Visual Forensic Detection of Fabricated Bangla News Photocards using Deep Learning

Computer Science > Computer Vision and Pattern Recognition arXiv:2611.05403 (cs) [Submitted on 19 Oct 2026] Title: Visual Forensic Detection of Fabricated Bangla News Photocards using Deep Learning Authors: Anindya Kundu, Kabya Mithun Saha, Md. Muhaiminul Islam, Fayek Ahmed, Aminul Kader Bulbul, Muhammad Ibrahim.

计算机科学 > 计算机视觉与模式识别 arXiv:2611.05403 (cs) [提交于 2026 年 10 月 19 日] 标题：利用深度学习进行孟加拉语虚假新闻图片卡的视觉取证检测 作者：Anindya Kundu, Kabya Mithun Saha, Md. Muhaiminul Islam, Fayek Ahmed, Aminul Kader Bulbul, Muhammad Ibrahim。

Abstract: In Bangladesh, much localized misinformation spreads as photocards: single-image news summaries that imitate the branding of well-known news organizations. Because the misleading message lives in the pixels rather than in machine-readable text, text-based fact-checking tools cannot see it. We recast photocard fake-news detection as visual image forgery detection.

摘要：在孟加拉国，许多本土化的虚假信息以图片卡的形式传播：即模仿知名新闻机构品牌形象的单图新闻摘要。由于误导性信息存在于像素中而非机器可读的文本中，基于文本的事实核查工具无法识别。我们将图片卡虚假新闻检测重新定义为视觉图像伪造检测。

We build a dataset of 969 authentic and 1,077 fabricated Bangla photocards scraped from public Facebook pages: authentic cards from the verified pages of seventeen national television broadcasters, fabricated ones from pseudo-broadcaster pages, anonymous profiles, and public groups. Sixteen photometric and morphological transformations per image grow the corpus to 32,736 samples.

我们构建了一个包含 969 张真实和 1,077 张虚假孟加拉语图片卡的数据集，这些图片均抓取自公开的 Facebook 页面：真实图片卡来自 17 家国家电视台的认证页面，虚假图片卡则来自伪广播机构页面、匿名个人资料和公共群组。通过对每张图像进行 16 种光度学和形态学变换，语料库规模扩大至 32,736 个样本。

A fully connected ANN reaches only about 52% accuracy, its loss stuck at ln(2) ~ 0.693, showing that flattening the image destroys the spatial cues forgery detection needs. A compact, fully batch-normalized four-block CNN with 489,954 parameters, about 1,644x smaller than the ANN, is evaluated with leakage-free stratified group five-fold cross-validation, which keeps every augmented version of a card in one fold.

一个全连接人工神经网络（ANN）的准确率仅达到约 52%，其损失值停滞在 ln(2) ≈ 0.693，这表明将图像扁平化处理会破坏伪造检测所需的空间线索。我们评估了一个紧凑的、经过完全批归一化的四模块卷积神经网络（CNN），其参数量为 489,954 个，比 ANN 小约 1,644 倍，并采用了无泄漏的分层组五折交叉验证，确保同一张卡的每个增强版本都保留在同一个折叠中。

Under this protocol the CNN reaches 87.45% +/- 0.53% pooled accuracy from scratch and 88.75% +/- 1.55% when its classifier head is warm-started from the ANN. An ImageNet-pretrained YOLOv8n-cls backbone (1.44M parameters) fine-tuned on the same folds reaches 97.36% +/- 0.42%, marking the headroom above the compact model.

在此协议下，该 CNN 从零开始训练的汇总准确率为 87.45% +/- 0.53%，当其分类器头部通过 ANN 进行热启动时，准确率达到 88.75% +/- 1.55%。在相同折叠上微调的 ImageNet 预训练 YOLOv8n-cls 主干网络（1.44M 参数）达到了 97.36% +/- 0.42% 的准确率，标志着该模型相较于紧凑型模型仍有提升空间。

We report per-class precision, recall, F1, and pooled confusion matrices, diagnose and fix a one-class collapse caused by an unnormalized classifier head, and give a manual error analysis and bias-variance diagnosis showing that the compact CNNs under-fit.

我们报告了各类的精确率、召回率、F1 分数和汇总混淆矩阵，诊断并修复了由未归一化的分类器头部导致的单类崩溃问题，并提供了手动错误分析和偏差-方差诊断，表明紧凑型 CNN 存在欠拟合现象。