🐛 Fixed #

• Unit Parsing: Improved logic for separating unit prices from product text, specifically handling cases where unit information might appear as trailing text or between other tokens more reliably.

🛠️ Changed #

• Bounding Box Stability: Added a default zero-degree angle to bounding boxes generated from multi-line entities, ensuring more consistent geometry during receipt reconstruction.

✅ Tests #

• DM Integration: Expanded assertions for dm-drogerie markt receipts to include detailed validation of unit quantities and unit prices for line items.

✨ Added #

• Aldi-style unit support: Added logic to detect and parse unit information (quantity/price) when printed on the line above the product name (common on Aldi receipts).
• Expanded Currency Support: Improved monetary regex to handle currency symbols appearing after the amount (e.g., 0,49 €).

🛠️ Changed #

• Skew Angle Optimization: Implemented a more robust, weighted-median approach for skew estimation across multiple frames, significantly improving alignment stability during video scans.
• Parsing Bounds: Refined rightBound and centerBound calculations (3/4 and 1/2 of receipt width respectively) to better distinguish between item amounts and unknown metadata.
• Unit Detection: Enhanced _tryParseUnit and _tryParseUnknown to use more precise horizontal center-bounds, preventing misclassification of product text as unknown entities.
• Integer Parsing: Improved integer extraction logic to ensure quantity detection doesn't accidentally pick up parts of decimal numbers.

✅ Tests #

• ALDI Integration: Added comprehensive integration tests for ALDI receipts, validating store name, total, purchase date, and multi-line unit/quantity parsing.

✨ Added #

• New store support: Added Kaufland and dm store support (including improved/real-world handling).
• Parsing support: Added handling for asterisk-marked receipt lines.

🛠️ Changed #

• Normalization: Normalize spaces more consistently to improve matching under OCR spacing glitches.
• Total-label handling: Refined label-related conditions to improve robustness.
• Position creation:
  • Prefer creating positions from the same row when possible.
  • Apply non-strict position creation only for unused amounts to avoid accidental matches.
• Build/CI hygiene: Ignore other build paths to reduce noise and avoid unintended file interactions.
• Internal: Dynamic script language adjustments and minor method renaming/cleanup.

✅ Tests #

• Added and reorganized tests, including expanded integration test assertions (e.g., more unit/quantity/price checks).

✨ Added #

• Extended integration test coverage: Added some comprehensive integration tests with special validation for deposit positions.

🐛 Fixed #

• Total identification: Fixed a critical bug where totals were not correctly identified when total labels were present.
• Unit parsing tolerance: Adjusted vertical tolerance for better multi-line receipt handling.

✨ Added #

• Extended integration tests: Added comprehensive integration tests for multiple receipt types (EDEKA, LIDL, and ALDI) with validation of store names, totals, purchase dates, positions, and specific line items.
• Text processor testing support: Added debugRunSynchronouslyForTests flag to ReceiptTextProcessor for synchronous parsing in test environments.
• Truncated frequency calculation: New calculateTruncatedFrequency method that merges truncated leading-token alternatives into their longer counterparts before counting frequency.

🛠️ Changed #

• Unified unit parsing: Merged _tryParseUnitQuantity and _tryParseUnitPrice into a single _tryParseUnit method for more efficient unit detection.
• Total label detection improvements:
  • Now uses normalized keys for better matching accuracy
  • Added prefix matching for total labels (e.g., "SUMME EUR" will match "summe")
• Parser refinements:
  • Fixed bug where _findTotalLabel was incorrectly called instead of _findTotal when searching for totals
  • Improved unit text extraction to avoid including numeric portions
  • Enhanced unknown entity filtering to prevent lines with too many numeric characters from being classified as unknown
• Normalization improvements:
  • Updated frequency calculation for alternative postfix texts to use truncated frequency method.
  • Moved truncated frequency calculator to the correct normalization method for alternative texts.
  • Refined alternative text frequency calculation in RecognizedProduct.
• Parser simplification: Reduced complexity by removing overly complicated logic and making the codebase simpler and more maintainable.
• OCR autocorrection: Reduced OCR autocorrection aggressiveness for better accuracy.
• Skew calculation: Removed outliers from skew angle calculation for more stable alignment.
• Ignored keywords: Added "Subtotal" to the list of keywords to ignore during product recognition.
• Timeout handling: Improved timeout behavior for better responsiveness.

🐛 Fixed #

• Entity type preservation: Fixed issue where entities were being incorrectly cast, now properly creates new instances with correct types when filtering (e.g., RecognizedTotal → RecognizedAmount, RecognizedTotalLabel → RecognizedUnknown)
• Unit boundary detection: Changed unit parsing to use _rightL(line) instead of _leftL(line) for proper right-bound checking
• Unit parsing: Fixed a critical bug where unit prices with negative total prices were not correctly handled with proper sign adjustment.
• Unit validation: Corrected unit price-quantity validation logic to properly check tolerance thresholds.
• Parsing stability: Multiple fixes to improve parsing reliability and reduce edge-case errors.

🛠️ Changed #

• Product recognition threshold: Increased strictness for position confirmation – now requires 80% of positions to pass validation (up from 75%) for more reliable receipt parsing.
• Product name handling: Improved handling of leading digits – now allows product names with leading digits if they contain 6 or more letters total, reducing false rejections.

🐛 Fixed #

• Normalization: Removed verbose debug comments from OCR correction logic, improving code clarity.
• Integration tests: Corrected asset path reference in integration tests.

🧹 Housekeeping #

• Removed .metadata tracking file from version control in example app.
• Updated .gitignore to exclude metadata files going forward.

✨ Added #

• Added Spar to the default store list.

🛠️ Changed #

• Store and total-label detection is now more tolerant to OCR spacing glitches.
• Store extraction now only considers candidates above the first amount line, reducing false positives.
• Improved skew angle estimation for more stable item alignment.
• Parser now performs a fallback pass to assign unmatched amounts, reducing missing line items.

🐛 Fixed #

• Purchase date handling no longer locks onto early, unstable detections; continuous scans refine it correctly.
• Minor stability improvements in the optimizer and parser.

✨ Added #

• Unit price & quantity parsing: extract quantity and price-per-unit for line items.
• allowedProductGroups option to control which product groups are accepted during parsing.
• Position bounding box exposure for recognized positions.
• Minor improvements around pseudo positions for better preview/UX.

🛠️ Changed #

• Tuned defaults for stability and performance (cache size, thresholds, timeouts).
• Normalization tweaks: treat + as a normal character; refined space handling; extended patterns.
• Removed legacy keyword groups (discount/deposit) in favor of allowedProductGroups.
• Internal type/use cleanups (prefer double/int where appropriate).
• Purchase date recognition is tuned for better robustness.

🐛 Fixed #

• Lost groups during merging: fixed issues where groups could disappear or be dropped in long/unstable scans.
• Deposit handling: corrected parsing so deposit lines don’t corrupt totals or item classification.
• Product group classification: fixes to ensure correct group assignment and consistent downstream behavior.
• General stability fixes across optimizer and parser components.

✨ Added #

• Unit tests covering additional normalization scenarios.

🛠️ Changed #

• Adjusted onScanTimeout signature in ReceiptRecognizer.
• Improved normalization and output: use normalized text consistently and refined purchase date output.
• Tuned optimizer cache for better stability and responsiveness.

🐛 Fixed #

• Corrected skew angle handling and bounds calculation for accurate receipt alignment.
• Prevented misclassification of prices as product names.

🐛 Fixed #

• Pass static analysis: Reformatted receipt_recognition-dart to reach 50/50 points

✨ Added #

• Integration test for single-image recognition in the example app:
  • example/integration_test/receipt_recognizer_test.dart
• purchaseDate detection and exposure on RecognizedReceipt.
• Configurability via the new options/tuning system:
  • ReceiptOptions to control parsing, validation, and scanning behavior.
  • ReceiptTuning for adjustable merge/stability thresholds.
  • kReceiptDefaultOptions as a sensible baseline.
  • ReceiptRuntime helper to apply options during parsing.

🛠️ Changed #

• processImage now always returns a RecognizedReceipt (never null), even when incomplete or invalid.
• Clarified result semantics for continuous scanning (video):
  • isValid: essential structure present and totals are coherent.
  • isConfirmed: result stabilized across frames (finalized by optimizer).
• Scanning guidance:
  • Video: stop when receipt.isValid && receipt.isConfirmed.
  • Single image: isConfirmed typically not achievable; treat isValid as sufficient or prompt re-capture.
• Parser and Optimizer refactored and improved for more robust extraction and more stable results across frames.

🧹 Deprecated #

• minValidScans in ReceiptRecognizer (constructor parameter) in favor of stability/confirmation rules and tuning via ReceiptOptions/ReceiptTuning.
• Legacy sum/sumLabel naming: prefer total (RecognizedTotal) and totalLabel (RecognizedTotalLabel).
• RecognizedCompany is considered legacy; prefer RecognizedStore.

🐛 Fixed #

• RecognizedGroup: Added type annotation to maxGroupSize property

🛠️ Changed #

• Updated to the most recent linter to avoid server side linting issues

✨ Added #

• Configurable stores support in the parsing pipeline, recognizer now accepts an options map that is forwarded to the receipt parser

• Example app updates:

  • Introduces opencv_dart in the example’s dependencies used for image preprocessing

🛠️ Changed #

• ReceiptRecognizer constructor: new optional options parameter
• Default scanning cadence: scanInterval increased from 50 ms → 100 ms

🐛 Fixed #

• ReceiptParser: Checking minValidScans on successful scan

🛠️ Changed #

• ReceiptNormalizer: Simplified normalization logic

🐛 Fixed #

• ReceiptRecognizer: Fixed the scan counter reset issue that was preventing proper scan completion
• iOS Deployment: Updated minimum iOS deployment target to 13.0 for compatibility

🛠️ Changed #

• ReceiptRecognizer: Increased default minValidScans from 3 to 5 for more reliable scanning

🐛 Fixed #

• Scan Validation: Fixed the issue where scans could complete prematurely before reaching the minimum valid scan count
• Match Percentage Calculation: Corrected calculation logic for sum match percentage to use actual values instead of scan-weighted values
• Ignore Keywords: Added "Subtotal" to the list of keywords to ignore during product recognition

🛠️ Changed #

• Product Display: Updated debug output to use normalized product text for better consistency
• Recognition Accuracy: Enhanced filtering of non-product lines to improve overall scan quality

📚 Documentation #

• Streamlined README to reflect current capabilities and scope
• Removed outdated roadmap/status references and aligned wording with the present feature set
• Clarified Implementation Status and reorganized sections for better readability
• Minor copyedits for consistency (terminology, headings, and formatting)

✨ Added #

• Stable Sum Detection: The scanner now identifies the most likely total amount by combining confidence and position tracking across multiple frames
• Live Bounding Boxes: Detected receipt items are now outlined live on the camera preview, making it easier to see what's recognized in real time

🛠️ Changed #

• Metadata Filtering: The engine is better at ignoring irrelevant lines like footer notes or fiscal metadata, keeping the product list clean
• Enhanced Merge Logic: Optimized how multi-frame receipts are combined, improving structure and reliability even in fragmented scans

🐛 Fixed #

• Incorrect Totals: Fixed an issue where prices from unrelated lines could falsely be interpreted as the receipt’s sum
• Layout Glitches: Resolved occasional misalignments or line skips when processing longer or poorly lit receipts

✨ Added #

• Smarter Scan Confidence Filtering: The recognition engine can now automatically ignore low-confidence items that might distort the total sum
• Status Getter: Added a simple flag to detect if scanning is still in progress
• Improved Example App: The example app now includes best practice instructions, live scan overlays, and a clearer UI for scanning and viewing receipts

🛠️ Changed #

• Internal Optimizations: Minor refactoring and documentation improvements to ensure better readability and maintainability of the code

🐛 Fixed #

• Sum Calculation Accuracy: Fixed cases where incorrect items could cause mismatches with the printed total
• Scan Stability: Improved reliability of the scan process for varied receipt lengths and layouts

✨ Added #

• Manual Receipt Acceptance: Added acceptReceipt() method to manually accept receipts with validation discrepancies
• Enhanced Validation Flow: Improved receipt validation with configurable thresholds and detailed status reporting
• Validation State Tracking: Added comprehensive validation states (complete, nearlyComplete, incomplete, invalid)
• Documentation: Added detailed documentation on validation workflow and manual acceptance

🛠️ Changed #

• Implementation Status Table: Improved formatting of status table in documentation
• Progress Reporting: Enhanced ScanProgress model with more detailed validation information
• Validation Logic: Refined match percentage calculation between calculated and declared sums

🐛 Fixed #

• Receipt Merging: Fixed edge cases in receipt merging that could lead to inaccurate total sums
• Stability Thresholds: Adjusted default stability thresholds for more reliable scanning

🐛 Fixed #

• Pass static analysis: Added curly braces to reach 50/50 points

✨ Added #

• Intelligent product name normalization via ReceiptNormalizer.normalizeFromGroup
• Graph-based item ordering using PositionGraph with fuzzy matching and topological sort
• CachedReceipt consolidation with improved trust-based merging logic
• Parallelized parsing and graph resolution using Dart's compute() isolates
• Outlier filtering for OCR noise and trailing receipt lines
• Unit tests for name cleaning and group resolution

🛠️ Changed #

• Restructured core/ folder for better modularity
• Improved ReceiptParser._shrinkEntities() to robustly remove lines below sum/total
• Enhanced debug output when scanning receipts in ReceiptRecognizer

📚 Documentation #

• Added concise DartDoc to all public APIs, data models, and helpers

✅ Tests #

• Added group-based name cleaning tests
• Validated space vs. no-space consensus logic (e.g. "Co ke Zero" vs. "Coke Zero")

📚 Documentation #

• Added comprehensive dartdoc comments for all model classes in lib/src/receipt_models.dart

✨ Added #

• Optimizer Module: Introduced a new optimizer module to enhance the performance and efficiency of the receipt recognition pipeline
• Product Name Detection Enhancements: Implemented advanced techniques for more accurate extraction and recognition of product names from receipts

🛠️ Changed #

• Company Regex Enhancements: Updated regular expressions related to company name detection to enhance parsing accuracy

🐛 Fixed #

• Product Name Extraction: Resolved issues where certain product names were not being accurately extracted due to inconsistent formatting
• Company Name Detection: Fixed bugs in the regular expressions that led to incorrect company name identification in specific receipt formats
• Parser Stability: Addressed edge cases in the parser that previously caused errors when processing receipts with uncommon layouts

✅ Added #

• Unit Tests: Introduced unit tests to enhance code reliability and ensure consistent behavior across updates

🛠️ Changed #

• Parser Refactoring:
  • Moved parsing methods to a dedicated parser module, improving code organization and maintainability
  • Converted parser methods to static, facilitating easier access without instantiating classes

🐛 Fixed #

• Company Name Recognition: Refined regular expressions related to company name detection, enhancing accuracy in identifying company names from receipts

✨ Added #

• ReceiptWidget: A customizable Flutter widget that displays parsed receipt data in a layout resembling a real supermarket receipt
• Advanced example demonstrating integration with a live video feed using the camera package

🛠️ Changed #

• Updated README.md to include usage examples and details about models etc.

🐛 Fixed #

• Minor bug fixes and performance improvements

✨ Added #

• Initial release of receipt_recognition.
• Core functionality to process images of receipts and extract structured data such as store name, items, and total amount using Google's ML Kit
• Support for both Android and iOS platforms