Replace gap-based splitting with alignment-bin approach: cluster word
left-edges within 8px tolerance, find the leftmost bin with >= 10% of
words as the true column start, split off any words to its left as a
sub-column. This correctly handles both page references ("p.59") and
misread exclamation marks ("!" → "I") even when the pixel gap is small.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Detects hidden sub-columns (e.g. page references like "p.59") within
already-recognized columns by clustering word left-edge positions and
splitting when a clear minority cluster exists. The sub-column is then
classified as page_ref and mapped to VocabRow.source_page.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Gaps that extend to the image boundary (top/bottom edge) are not valid
content separators — they typically represent dewarp padding. Only gaps
with content on both sides qualify as header/footer boundaries.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The inverted image can be taller than img_h after dewarp shear
correction, causing footer_y to be detected outside the visible page.
Now clamps the horizontal projection to actual_h = min(inv.shape[0], img_h).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Check for actual ink content in detected top/bottom regions:
- 'header'/'footer' when text is present (e.g. title, page number)
- 'margin_top'/'margin_bottom' when the region is empty page margin
Also update all skip-type sets and color maps for the new types.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Replace the trivial top_y/bottom_y threshold check with horizontal
projection gap analysis that finds large whitespace gaps separating
header/footer content from the main body. This correctly detects
headers (e.g. "VOCABULARY" banners) and footers (page numbers) even
when _find_content_bounds includes them in the content area.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The ocr_pipeline_api.py code path called classify_column_types without
left_x/right_x, so margin regions were never created. Also add logging
to _build_margin_regions for debugging.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Thin black lines (1-5px) at page edges from scanning were incorrectly
detected as content, shifting content bounds and creating spurious
IGNORE columns. This filters narrow projection runs (<1% of image
dimension) and introduces explicit margin_left/margin_right regions
for downstream page reconstruction.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
1. Unit tests: 76 new parametrized tests for noise filter, phonetic detection,
cell text cleaning, and row merging (116 total, all green)
2. Continuation-row merge: detect multi-line vocab entries where text wraps
(lowercase EN + empty DE) and merge into previous entry
3. Empty DE fallback: secondary PSM=7 OCR pass for cells missed by PSM=6
4. Batch-OCR: collect empty cells per column, run single Tesseract call on
column strip instead of per-cell (~66% fewer calls for 3+ empty cells)
5. StepReconstruction UI: font scaling via naturalHeight, empty EN/DE field
highlighting, undo/redo (Ctrl+Z), per-cell reset button
6. Session reprocess: POST /sessions/{id}/reprocess endpoint to re-run from
any step, with reprocess button on completed pipeline steps
Also fixes pre-existing dewarp_image tuple unpacking bug in run_cv_pipeline
and updates dewarp tests to match current (image, info) return signature.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Two fixes:
1. Tokens ending with ] (e.g. "serva]") were stripped by the noise
filter because ] was not in the allowed punctuation list.
2. Rows containing only phonetic transcription (e.g. ['mani serva])
are now merged into the previous vocab entry instead of creating
a separate (invalid) entry. This prevents the LLM from trying
to "correct" phonetic fragments.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The noise filter was stripping words containing hyphens, parentheses,
slashes, and dots (e.g. "money-saver", "Schild(chen)", "(Salat-)Gurke",
"Tanz(veranstaltung)"). Now strips all common dictionary punctuation
before checking for internal noise characters.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Replace containment-with-padding approach with midpoint-based column
ranges. For adjacent columns, the assignment boundary is the midpoint
between them (Voronoi-style). This prevents padding overlap where words
near column borders (e.g. "We" at the start of example sentences) were
assigned to the preceding column. The last column extends generously to
capture all rightmost text.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
_is_noise_tail_token() treated words with unbalanced parentheses like
"selbst)" or "(wir" as OCR noise because the parenthesis counted as
"internal noise". Now strips leading/trailing parentheses before the
noise check, so legitimate words in example sentences like
"We baked ... (wir ... selbst)" are preserved.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Word assignment: Replace nearest-center-distance with containment-first
strategy. Words whose center falls within a column's bounds (+ 15% pad)
are assigned to that column before falling back to nearest-center. This
fixes long example sentences losing their rightmost words to adjacent
columns.
LLM review: Strengthen prompt to explicitly forbid changing proper nouns,
place names, and correctly-spelled words. Add _is_spurious_change()
post-filter that rejects case-only changes and hallucinated word
replacements (< 50% character overlap).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
StepLlmReview: Show full vocab table with image overlay, row-level
status tracking (pending/active/reviewed/corrected/skipped), and
auto-scroll during SSE streaming. Load previous results on mount.
StepReconstruction: New step 7 with editable text fields at original
bbox positions over dewarped image. Zoom controls, tab navigation,
color-coded columns, save to backend.
Backend: Add POST /sessions/{id}/reconstruction endpoint.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Stream LLM review results batch-by-batch (8 entries per batch) via SSE
- Frontend shows live progress bar, batch log, and corrections appearing
- Skip entries with IPA phonetic transcriptions (already dictionary-corrected)
- Refactor llm_review_entries into reusable helpers for both streaming and non-streaming paths
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add /no_think tag to prompt (qwen3 thinking mode causes massive slowdown)
- Increase httpx timeout from 120s to 300s for large vocab tables
- Improve error logging with traceback and exception type
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Replace the placeholder "Koordinaten" step with an LLM review step that
sends vocab entries to qwen3:30b-a3b via Ollama for OCR error correction
(e.g. "8en" → "Ben"). Teachers can review, accept/reject individual
corrections in a diff table before applying them.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add _KNOWN_ABBREVIATIONS set with ~150 common EN/DE abbreviations
(sth, sb, etc, eg, ie, usw, bzw, vgl, adj, adv, prep, sg, pl, ...).
Tokens matching known abbreviations are never stripped as noise.
Also handle dotted abbreviations (e.g., z.B., i.e.) that have no
2+ consecutive alpha chars by checking the abbreviation set before
the _RE_REAL_WORD filter.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add _clean_cell_text() with three sub-filters to remove OCR noise:
- _is_garbage_text(): vowel/consonant ratio check for phantom row garbage
- _is_noise_tail_token(): dictionary-based trailing noise detection
- _RE_REAL_WORD check for cells with no real words (just fragments)
Handles balanced parentheses "(auf)" and trailing hyphens "under-"
as legitimate tokens while stripping noise like "Es)", "3", "ee", "B".
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Two generic noise filters added to _ocr_single_cell():
1. Word confidence filter (conf < 30): removes low-confidence words
before text assembly. Catches trailing artifacts like "Es)" after
real text, and standalone noise from image edges.
2. Cell noise filter: clears cells whose entire text has no real
alphabetic word (>= 2 letters). Catches fragments like "E:", "3",
"u", "D", "2.77", "and )" from image areas, while keeping real
short words like "Ei", "go", "an".
Both filters apply to word-lookup AND cell-OCR fallback results.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The cell grid IS the result. Each cell stays at its detected position.
Removed _split_comma_entries and _attach_example_sentences from the
pipeline — they were shuffling content between rows/columns, causing
"Mäuse" to appear in a separate row, "stand..." to move to Example,
and "Ei" to disappear.
Now: cells → _cells_to_vocab_entries (1:1 row mapping) →
_fix_character_confusion → _fix_phonetic_brackets → done.
Also lowered pixel-density threshold from 2% to 0.5% for the cell-OCR
fallback so small text like "Ei" is not filtered out.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Three non-generic solutions replaced with universal heuristics:
1. Cell-OCR fallback: instead of restricting to column_en/column_de,
now checks pixel density (>2% dark pixels) for ANY column type.
Truly empty cells are skipped without running Tesseract.
2. Example-sentence detection: instead of checking for example-column
text (worksheet-specific), now uses sentence heuristics (>=4 words
or ends with sentence punctuation). Short EN text without DE is
kept as a vocab entry (OCR may have missed the translation).
3. Comma-split: re-enabled with singular/plural detection. Pairs like
"mouse, mice" / "Maus, Mäuse" are kept together. Verb forms like
"break, broke, broken" are still split into individual entries.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Three bugs in the post-processing pipeline were overwriting correct
streaming results with wrong ones:
1. _split_comma_entries was splitting "Maus, Mäuse" into two separate
entries. Disabled — word forms belong together.
2. _attach_example_sentences treated "Ei" (2 chars) as OCR noise due
to `len(de) > 2` threshold. Lowered to `len(de) > 1`.
3. _attach_example_sentences wrongly classified rows with EN text but
no DE (like "stand ...") as example sentences, merging them into
the previous entry. Now only treats rows as examples if they also
have no text in the example column.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Skip Tesseract fallback for column_example cells which are often
legitimately empty. This reduces ~48 Tesseract calls to ~10,
cutting Step 5 fallback time from ~13s to ~3s.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Next.js was producing the same chunk hash across builds, causing
browsers to serve stale cached JS even after redeployment.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Word-lookup from full-page Tesseract is fast but can miss small or
isolated words (e.g. "Ei"). Now falls back to per-cell Tesseract OCR
for cells that remain empty after word-lookup. The ocr_engine field
reports 'cell_ocr_fallback' for cells that needed the fallback.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Word-lookup is now ~0.03s (vs seconds with per-cell Tesseract), so
always re-run detection when entering Step 5 instead of showing
potentially stale cached word_result from the session DB.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Replace per-cell word filtering (which allowed the same word to appear in
multiple columns due to padded overlap) with exclusive nearest-center
assignment. Each word is assigned to exactly one column per row.
Also use row height as Y-tolerance for text assembly so words within
the same row (e.g. "Maus, Mäuse") are always grouped on one line.
Fixes: words leaking into wrong columns, missing words, duplicate words.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The row_result stored in DB excludes words to keep payload small.
When Step 5 reconstructs RowGeometry from DB, words were empty,
causing word-lookup to find nothing and return blank cells.
Now re-populates row.words from cached _word_dicts (or re-runs
detect_column_geometry if cache is cold) before cell grid building.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Replace per-cell Tesseract re-runs with lookup of pre-existing full-page
words from row.words. Words are filtered by X-overlap with column bounds.
This fixes phantom rows with garbage text, missing last words, and
incomplete example text by using the more reliable full-page OCR results.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Rows in inter-line whitespace gaps have no Tesseract words assigned but
were still processed by build_cell_grid, producing garbage OCR output.
Filter these phantom rows using the word_count field set during Step 4.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Cells now appear one-by-one in the UI as they are OCR'd, with a live
progress bar, instead of waiting for the full result.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Extract build_cell_grid() as layout-agnostic foundation from
build_word_grid(). Step 5 now produces a generic cell grid (columns x
rows) and auto-detects whether vocab layout is present. Frontend
dynamically switches between vocab table (EN/DE/Example) and generic
cell table based on layout type.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The validation that rejected word-center grid when it produced more rows
than gap-based detection was causing fallback to gap-based rows (large
boxes). The word-center grid regularization works correctly after the
center-based grouping and cluster merging fixes.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Two fixes:
1. Grid validation: reject word-center grid if it produces MORE rows
than gap-based detection (more rows = lines were split = worse).
Falls back to gap-based rows in that case.
2. Words overlay: draw clean grid cells (column × row intersections)
instead of padded entry bboxes. Eliminates confusing double lines.
OCR text labels are placed inside the grid cells directly.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The y_tolerance for word-center clustering was based on median word
height (21px → 12px tolerance), which was too small. Words on the
same line can have centers 15-20px apart due to different heights.
Now uses 40% of the gap-based median row height as tolerance (e.g.
40px row → 16px tolerance), and 30% for merge threshold. This
produces correct cluster counts matching actual text lines.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
When columns change (Step 3), invalidate row_result and word_result.
When rows change (Step 4), invalidate word_result.
This ensures Step 5 always uses the latest row boundaries instead of
showing stale cached word_result from a previous run.
Applies to both auto-detection and manual override endpoints.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Fix half-height rows caused by tall special characters (brackets, IPA
symbols) being split into separate line clusters:
- Group words by vertical CENTER instead of TOP position, so tall
characters on the same line stay in one cluster
- Filter outlier-height words (>2× median) when computing letter_h
so brackets/IPA don't skew the row height
- Merge clusters closer than 0.4× median word height (definitely
same text line despite slight center differences)
- Increased y_tolerance from 0.5× to 0.6× median word height
- Enhanced logging with cluster merge count and row height range
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Replace rigid uniform grid with bottom-up approach that derives row
boundaries from word vertical centers:
- Group words into line clusters, compute center_y per cluster
- Compute pitch (distance between consecutive centers)
- Detect section breaks where gap > 1.8× median pitch
- Place row boundaries at midpoints between consecutive centers
- Per-section local pitch adapts to heading/paragraph spacing
- Validate ≥85% word placement, fallback to gap-based rows
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Replace _split_oversized_rows() with _regularize_row_grid(). When ≥60%
of content rows have consistent height (±25% of median), overlay a
uniform grid with the standard row height over the entire content area.
This leverages the fact that books/vocab lists use constant row heights.
Validates grid by checking ≥85% of words land in a grid row. Falls back
to gap-based rows if heights are too irregular or words don't fit.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Implement _split_oversized_rows() in detect_row_geometry() (Step 7) to
split content rows >1.5× median height using local horizontal projection.
This produces correctly-sized rows before word OCR runs, instead of
working around the issue in Step 5 with sub-cell splitting hacks.
Removed Step 5 workarounds: _split_oversized_entries(), sub-cell
splitting in build_word_grid(), and median_row_h calculation.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
For cells taller than 1.5× median row height, split vertically into
sub-cells and OCR each separately. This fixes RapidOCR losing text
at the bottom of tall cells (e.g. "floor/Fußboden" below "egg/Ei"
in a merged row). Generic fix — works for any oversized cell.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Adds entries for all regulation codes in REGULATIONS_IN_RAG that were
missing from RAG_PDF_MAPPING, fixing "Kein PDF-Mapping" messages.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Integrate Britfone dictionary (MIT, 15k British English IPA entries)
- Add pronunciation parameter: 'british' (default) or 'american'
- British uses Britfone (Received Pronunciation), falls back to CMU
- American uses eng_to_ipa/CMU, falls back to Britfone
- Frontend: dropdown to switch pronunciation, default = British
- API: ?pronunciation=british|american query parameter
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The Docker container cannot reach Google Fonts, causing build failures.
Switch to bundled local font file using next/font/local.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
1. Semantic example matching: instead of attaching example sentences
to the immediately preceding entry, find the vocab entry whose
English word(s) appear in the example. "a broken arm" → matches
"broken" via word overlap, not "egg/Ei". Uses stem matching for
word form variants (break/broken share stem "bro").
2. Cell padding: add 8px padding to each cell region so words at
column/row edges don't get clipped by OCR (fixes "er wollte"
missing at cell boundaries).
3. Treat very short DE text (≤2 chars) as OCR noise, not real
translation — prevents false positives in example detection.
All fixes are generic and deterministic.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
