Coloring in the world of others: color use in visual orientalism, 1890-1920

The paper examines how color media shaped historical color sense and contributed to the cultural production of the 'Orient' and 'Occident' around 1890–1920. It situates the work within debates about whether color perception and naming are universal and evolutionary (Berlin & Kay) or culturally conditioned and utility-driven (Eco; subsequent work on communicative needs). The authors argue that globalization and a late-19th-century 'color revolution' mediatized color perception through new dyes and color image media, making color sense increasingly shaped by media affordances. They highlight methodological challenges for studying historical color (aging of materials, digitization altering colors, and shifting mappings from color space to terms). Research questions: (1) Do specific media affect color use? (2) Can dominant colors distinguish representations of the Orient and Occident across media? (3) Does the presence or absence of specific colors define visual Orientalism? The study leverages machine learning on large digitized collections to bridge color perception research with historical visual representation.

Building on Said’s theory of Orientalism, the paper reviews visual Orientalism in painting and photography. Nochlin showed how 19th-century orientalist painters constructed an exoticized, static, and backward 'Orient' to mirror Western identity. Photography, while ostensibly objective, often reproduced and stabilized the same tropes, especially via mass media like postcards that democratized access to Orientalist imagery. Color was central to the 'Orientalist aesthetic'—artists sought to render the 'brilliant colors' of the Near and Middle East. The arrival of photochrome (hand-colored, mass reproducible) and autochrome (color captured via plate and light) offered new color technologies. Photochromes, despite claims to depict 'colors of nature,' reflected printers’ subjective palette choices. Autochromes were advertised as capturing faithful colors, though their creators still worked within orientalist representational traditions. The two media had different uses and audiences: photochromes proliferated via tourism markets; autochromes were harder to reproduce and mainly used by artists and amateurs but were also exhibited.

Data: Two large digitized collections were analyzed—approximately 6,500 photochromes (1890–1910, from Photoglob and Detroit Publishing Company via Library of Congress) and about 65,000 autochromes (1909–1931, Albert Kahn Archives de la Planète). Countries were mapped to 'Orient' and 'Occident' for analysis; France was excluded from Occident due to overrepresentation in autochromes. This yielded 6,685 Occident and 7,024 Orient autochromes, and 4,435 Occident and 371 Orient photochromes. Digitization and preprocessing: Photochromes were scanned in 2001–2002 at 400 ppi; Kahn autochromes were digitized locally (circa 90 images/day). While color charts were likely used, none are visible in the files. To mitigate scanning and aging differences, a clamping method normalized brightness. Uniform cropping removed borders and watermarks (photochromes: 0.85 crop to remove white border; autochromes: bottom 0.75, and 0.92 on left/top/right to remove watermark and textual paper areas). Geographic names were aligned to accommodate historical/modern variants. Feature extraction and modeling steps: 1) K-means clustering extracted 8 and 16 dominant colors (in RGB) per image. 2) Relative frequencies of these colors provided weights. 3) Colors and weights were placed into equally sized RGB 'color buckets' by dividing each RGB dimension into 4, 6, or 8 parts, yielding n_division^3 buckets (e.g., 64, 216, 512). 4) Random forest classifiers used bucketed features to: (A) distinguish autochrome vs photochrome; (B) classify Orient vs Occident within photochromes; (C) classify Orient vs Occident within autochromes. 5) SHAP (Shapley Additive Explanations) interpreted feature impacts at the level of specific color buckets. The study used RGB (not perceptually uniform spaces like CIELAB) because the aim was to test whether high-level bucketed color features can differentiate categories rather than model precise perception. Evaluation: Accuracy and F1 scores were reported across configurations varying number of dominant colors (8, 16) and number of buckets (64, 216, 512).

- Media differentiation: Dominant colors robustly distinguished media type. The autochrome vs photochrome classifier reached up to 0.95 accuracy and 0.95 F1 (16 colors, 512 buckets). Accuracy was high across settings (0.92–0.95), empirically demonstrating that the two media impose distinct color palettes and that color sense became mediated by technology around 1900. - Orient vs Occident in photochromes: The classifier achieved high performance (up to 0.93 accuracy and 0.93 F1 with 16 colors, 512 buckets; also strong at 0.92 with 16/216), indicating that printers’ palette decisions encoded a distinct 'Orientalist' vs 'Occidental' color usage. - Orient vs Occident in autochromes: Performance was much lower (best around 0.68 accuracy and 0.67 F1 with 16 colors, 512 buckets; generally 0.62–0.68), only modestly above chance. This suggests autochromes, whose colors derive from light during exposure, yielded more neutral or less stereotypically segregated color palettes for Orient/Occident. - SHAP insights: For media classification, the absence of certain buckets (e.g., 146, 438, 365, 73) predicted autochromes, while absence of bucket 229 indicated photochromes, implying non-overlapping palette affordances between media. For photochromes’ Orient vs Occident, presence of bucket 447 favored 'Orient', and its absence favored 'Occident'; presence/absence of other buckets (e.g., 1 and 18) also shifted predictions, showing that both presence and absence of specific colors contributed to visual Orientalism. - Overall: Results support three conclusions: (1) late-19th-century mediatization of color sense; (2) in photochromes, specific colors’ presence/absence was central to visual Orientalism; (3) autochromes provided a more objective/neutral color rendering of the Near and Middle East than photochromes.

Findings show that media affordances significantly shaped historical color sense beyond linguistic factors. As globalization and new color technologies spread, color perception became mediated by reproducible palettes and printing processes. The strong separability of photochromes and autochromes confirms distinct media palettes, while the high separability of Orient vs Occident within photochromes indicates that printers’ subjective color choices encoded Orientalist aesthetics as formal color differences, not only content tropes. Conversely, autochromes’ lower separability implies that capturing color from light constrained or diluted the ability to sustain an imagined color divide between Orient and Occident at the aesthetic level. The study broadens visual Orientalism research by foregrounding color as a formal mechanism for constructing 'imagined geographies' and highlights that perceived realism (vivid color photochromes) could mask fictitious, constructed color worlds. The authors advocate integrating media materiality into color perception studies, complementing linguistic and cultural explanations.

The study provides empirical evidence that: (1) color sense became mediated in the late 19th century through color image technologies; (2) in photochromes, the presence and absence of particular colors were key to visual Orientalism, differentiating Orient and Occident; and (3) autochromes, deriving color from light exposure, offered a more neutral or complex picture of the Near and Middle East than photochromes. The work opens avenues for tracing the mediatization of color across further technological transitions (e.g., Dufaycolor, Kodachrome, Agfacolor-NEU, Technicolor, digital color, and AI-generated imagery) to understand how evolving media palettes have shaped cultural color sense over the last two centuries.

- Historical color drift and digitization: Original colors may have shifted; early-2000s digitization workflows (scanner differences, lack of visible color charts) can introduce variation. Brightness clamping mitigates but cannot eliminate these effects. - Color space and features: Use of RGB and bucketed dominant colors prioritizes separability over perceptual uniformity (e.g., not using CIELAB); results reflect high-level palette differences rather than fine-grained perceptual distinctions. - Class imbalance and coverage: Photochromes have far fewer Orient images (371) than Occident (4,435), potentially affecting classifier stability; France was excluded from Occident in autochromes due to overrepresentation; country-to-category mapping is an analytical construct with fuzzy historical boundaries. - Content not modeled: The approach targets color as a formal feature; it does not capture semantic content (e.g., mosques vs mountains) that also conveys Orientalist tropes. - No close readings: The study intentionally avoids close readings; while aligning with its goals, it limits image-specific interpretive nuance. - Generalizability: Findings pertain to these two collections and digitization pipelines; broader claims across media, periods, and repositories require further validation.