Source image: WhatsApp Image 2026-09-05 at 07.26.13.jpeg (photo of a printed thesis abstract page). Transcribed text is at kathy/abstract.
Ran tesseract directly on the image, then again after preprocessing with ImageMagick (grayscale, normalize, sharpen, 2x resize) and --psm 6.
The photographed page has visible curvature. The two flatter middle paragraphs (springback and elongation prediction) OCR'd cleanly and matched the source image exactly. The top and bottom paragraphs, which fall on the more curved parts of the page, came out with scrambled word order.
Cross-checked the scrambled tesseract output against a direct visual read of the image. The garbled paragraphs contained the same key phrases as the visual reading (e.g. "company scheme", "structured scientific approach", "draw bending process", "results from the research ... graphs, with equations derived from them ... further work"), confirming no content was misread, only word order disrupted by the curved lines. The final text in kathy/abstract was assembled from the visual reading and cross-checked against tesseract's output.