Background and Scholarly Problem

Reconstructing how historical photographs were edited and repurposed in Chinese wartime magazines is a complex challenge. Archives like the  Jinchaji Pictorial (a Chinese Communist Party photographic magazine from the WWII era) contain thousands of images across issues, but they include no records of the editorial decisions behind those images. Historians must therefore infer editorial strategies and historical context from large, heterogeneous collections of photographs without direct documentation. Digital Historical Forensics (DHF) is introduced as a methodological bridge between traditional contextual analysis and computational techniques to address this problem. By integrating humanistic media research with computer vision, DHF allows us to systematically track how photographs circulated and transformed across publications, reconstructing the editorial logic behind these visual narratives.

Methodological Innovation

The DHF pipeline combines automated image processing with cross-domain image similarity analysis in an accessible way. First, an object detection model (YOLOv7) automatically crops individual photographs from digitized magazine pages. This step isolates each image, preparing it for comparison without requiring manual clipping. Next, a multi-model ensemble of state-of-the-art computer vision networks – specifically a Vision Transformer, EfficientNetV2, and Swin Transformer – is trained on the cropped images alongside a large dataset of historical photographs. This ensemble compares images across different sources to find matches, effectively spotting when a photograph from a photographer’s archive reappears in a magazine, even if it has been altered or recontextualized.

Using this pipeline, the system can detect instances of image reuse and editorial alteration at scale. Notably, the model achieved a top-15 retrieval accuracy of 77.8% in evaluations, meaning that for a given query image, the correct match (if it exists) is usually found among the top 15 results. Such performance demonstrates the robustness of our approach and its practical utility for examining large archival collections.

Figure 1: Diagram of the proposed computer vision pipeline for image retrieval.

Revealing Editorial Strategies (Findings)

Applying DHF to wartime magazine archives uncovers patterns of image reuse and manipulation that shed light on editorial strategies and ideological shifts over time. For example, the pipeline might find that a 1943 frontline photograph reappeared in a 1946 magazine issue with significant changes. In the later version, editors cropped the image to focus on a single heroic figure and paired it with a new caption that reframed its meaning. Such alterations demonstrate how a documentary scene was repackaged to serve a different narrative context.

Through these findings, we observe how editors preferred certain visual themes – often emphasizing human subjects or adding textual elements – in line with socialist realist aesthetics. Recurrent editorial interventions like cropping, retouching, and recaptioning frequently transformed images from straightforward records into metaphorical propaganda messages. Over time, tracking these changes reveals how photographs were continuously repurposed to align with shifting political agendas and narrative needs. In other words, magazine editors systematically turned war photographs into symbols, illustrating how visual evidence was refashioned to support evolving ideological goals.

Implications for Visual Culture and Media History

This project demonstrates how a computationally enhanced approach can expand interpretive possibilities in visual culture studies, media history, and historical epistemology. By resituating photographs in their original publication contexts and mapping their reuses, we treat images not as static artifacts but as dynamic entities with evolving “circulation lives.” The DHF approach effectively maps hidden networks of image dissemination and transformation, reframing historical imagery as a dynamic site of contested meaning rather than a fixed representation of reality. This contributes to media history by detailing how editorial practices in propaganda magazines shaped what readers saw and understood, turning photographs into evolving narratives rather than static evidence. In terms of historical epistemology, the work highlights how our knowledge of the past is mediated by editorial choices: the absence or presence of a caption, a cropped-out figure, or an altered background can profoundly change the story a photograph tells.

Ultimately, Digital Historical Forensics serves as a convergence of humanistic inquiry and machine learning techniques, enabling scholars to investigate large visual archives in ways previously impractical. Rather than replacing traditional analysis, this method augments it: researchers can quickly identify image reuse, trace editorial interventions, and then interpret these findings within a broader socio-historical context. The ability to computationally sift through vast collections for visual patterns means that questions about propaganda, editorial intent, and the evolution of visual rhetoric can be explored systematically and at scale. This poster showcases how combining close contextual analysis with computer vision not only reconstructs the editorial strategies behind Chinese wartime magazines, but also opens new avenues for understanding how images functioned as malleable documents-turned-metaphors in the construction of history.