How arbom/tablice works
1. Repository shape
- This is a single-process C# console application (Tablice) built on .NET Framework 4.7.2.
- The system is not split into services or tiers; instead, it is a local image-processing pipeline that runs end-to-end inside one executable.
2. Primary runtime boundary
- The app consumes a graphic file from the local filesystem and produces a recognized registration number.
- There are no web endpoints, databases, message queues, or cloud dependencies visible in the tree.
3. Top-level orchestration
- Program.cs is the execution entrypoint and likely coordinates the full recognition flow.
- It wires together detection, rectification, character extraction, and classification into one sequential pass.
4. Core external dependency
- The key library is EmguCV 4.4.0, which is the .NET wrapper around OpenCV.
- That means the architecture is centered on classic computer-vision operations: masking, contour analysis, connected components, perspective correction, and image segmentation.
5. Plate detection subsystem
- The first stage finds the plate region in the source image.
- According to the README, this uses blue and white masks, merges candidate regions, then searches for the largest connected area and its corners.
- The corner-finding logic appears decomposed into separate helpers: Helper LeftUpCorner.cs, Helper RightUpCorner.cs, Helper LeftDownCorner.cs, Helper RightDownCorner.cs.
6. Plate rectification subsystem
- Once the corners are found, the detected plate is isolated and reshaped/perspective-corrected into a normalized plate image.
- Plate.cs is likely the domain object for this stage, representing the extracted plate region and its transformed form.
- This stage is important because it converts a skewed real-world image into a canonical input for character reading.
7. Character segmentation subsystem
- After rectification, the app identifies individual character contours on the plate.
- Mark.cs likely represents one detected sign/character candidate, including its bounding area and features used for recognition.
- The output of this stage is a sequence of character regions, each isolated for classification.
8. Classification subsystem
- The final recognition step uses an SVM-based classifier.
- The Data/SVM.txt and Data/SVM.xml files look like persisted model/data artifacts used by the classifier at runtime.
- The classifier assigns labels to each segmented character, then the app concatenates them into the final plate number.
9. Supporting image-processing utilities
- Helper.cs is the shared utility layer for reusable CV operations.
- It likely contains common functions for thresholding, contour handling, coordinate math, and image conversion, while the four corner-specific helper files isolate directional logic.
10. Project assets and configuration
- App.config and packages.config support the .NET app/runtime package setup.
- Resources.* suggests bundled application resources, while Data/ holds model artifacts needed to run recognition locally.
- This makes the project self-contained: code plus trained recognition data live together in the same repository.
11. End-to-end data flow
- The flow is: input image → masking/region detection → corner detection → plate normalization → character contour extraction → SVM classification → assembled plate string.
- For an architecture diagram, the main boundaries to show are:
- image input,
- plate localization,
- geometric correction,
- glyph segmentation,
- SVM inference,
- final string output.
12. Implementation style
- The architecture is algorithm-centric, not framework-centric.
- Most complexity is in the vision pipeline and geometric reasoning, with a thin console shell around it.
Main entry points: Plate.cs Plate Find · Program.cs Program · SVM.xml SVM · Helper LeftUpCorner.cs Left Up
Components 16
Runtime
Console entrypoint and application shell
- Program · Console application entrypoint that orchestrates the recognition flow
Tablice/Program.cs - Plate Text · Final recognized registration number assembled from classified characters
Vision pipeline
Plate recognition stages from image input to output
- Left Up · Helper for detecting the top-left plate corner
Tablice/cs/Helper LeftUpCorner.cs - Right Up · Helper for detecting the top-right plate corner
Tablice/cs/Helper RightUpCorner.cs - Left Down · Helper for detecting the bottom-left plate corner
Tablice/cs/Helper LeftDownCorner.cs - Right Down · Helper for detecting the bottom-right plate corner
Tablice/cs/Helper RightDownCorner.cs - Plate Find · Locates candidate plate regions using masking and connected-area analysis
Tablice/cs/Plate.cs - Plate Corners · Combines corner detection results into a plate quadrilateral
Tablice/cs/Plate.cs - Plate Warp · Perspective-corrects the detected plate into a normalized image
Tablice/cs/Plate.cs - Mark · Represents a detected character or sign candidate with its features
Tablice/cs/Mark.cs - Segment · Extracts and orders character regions from the rectified plate image
Tablice/cs/Mark.cs - Features · Builds recognition features from segmented character candidates
Tablice/cs/Mark.cs
Support
Shared helpers, resources, and packaged model data
- Helper · Shared image-processing and geometry utilities used across stages
Tablice/cs/Helper.cs - SVM · Classifies character features using the persisted SVM model
Tablice/Data/SVM.xml - SVM Data · Serialized SVM training or metadata data consumed by the classifier
Tablice/Data/SVM.txt - Resources · Application resources bundled with the console app
Tablice/Properties/Resources.resx
Connections 19
- Program → Plate Find: starts · Entry point launches the recognition pipeline
- Program → Helper: uses · Coordinates shared CV helpers during processing
- Plate Find → Left Up: corner hint · Uses directional helper for the top-left corner
- Plate Find → Right Up: corner hint · Uses directional helper for the top-right corner
- Plate Find → Left Down: corner hint · Uses directional helper for the bottom-left corner
- Plate Find → Right Down: corner hint · Uses directional helper for the bottom-right corner
- Left Up → Plate Corners: feeds · Supplies one detected corner
- Right Up → Plate Corners: feeds · Supplies one detected corner
- Left Down → Plate Corners: feeds · Supplies one detected corner
- Right Down → Plate Corners: feeds · Supplies one detected corner
- Plate Corners → Plate Warp: normalizes · Defines the plate quadrilateral for perspective correction
- Plate Warp → Segment: produces · Provides a normalized plate image for segmentation
- Segment → Mark: creates · Wraps detected character regions as mark objects
- Mark → Features: describes · Transforms each candidate into classifier-ready features
- Features → SVM: classified by · SVM assigns a label to each feature vector
- SVM → SVM Data: loads · Uses persisted model data at runtime
- Plate Find → Resources: may use · May rely on bundled resources for processing or display
- SVM → Plate Text: assembles · Predicted character labels are concatenated into the final plate string
- Program → Plate Text: returns · Console app prints or emits the recognized result