Skip to content
Handwritten Math Expressions Recognition
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
BONUS
DataPreparation
Model sync May 11, 2017
Prediction
Report
Segmentation
output
.gitignore sync May 9, 2017
261_892_130_143.png
README.md
Report.pdf
test_resize.py
testouttt.png

README.md

HME_recognition

Handwritten Math Expressions Recognition

Roadmap Outline

Data Preparation

  • Refine on given training set
  • NIST – Refine (For English Characters)
  • CROHME (For math symbols): Online dataset to offline-like data : opencv.dilate + Gaussian blur

Model

Classifier with following symbols:

  • Digits: 0-9

  • Characters: x, y, a, b, c, d, m, n, p, delta, f, h, k, sin, cos, tan, A, pi

  • Operators: +, -, *, /, =, sqrt, ^, _, bar, frac, cdots, (, )

    Source Codeg

  • Architecture: ResNet-50

Segmentation

  • opencv.findContours()

    Source Code

  • Small window filters with various sizes to scan after opencv.findContours()

  • A Region Proposal-like Network for incompletely segmented parts

RNN - Context

Merge decomposition parts of signs like =, /, cdots to one rectangle with RNN(LSTM)

RNN

From symbol locations (4 points) and its label to adjust label and write to data structure

Data structures and helpers

  • Tree-like ADT to hold math expressions
  • From ADT to latex
You can’t perform that action at this time.