Deslanting Algorithm

• Update 2021/2: added Python package
• Update 2021/1: added Python implementation

This algorithm sets handwritten text in images upright, i.e. it removes the cursive writing style. One can use it as a preprocessing step for handwritten text recognition. The following illustration shows input and output of the algorithm for a given image (data/test1.png).

There are three implementations provided (Python, C++, OpenCL) with a focus on the Python implementation. For C++ and OpenCL see the folder extras.

Installation

• Go to the root level of the repository
• Install by running pip install .
• Run deslant_img (without arguments) from the command line to process the images in the data directory (images taken from IAM and Bentham dataset)
• This opens a window showing the input image, deslanted image and score values
• The script can be configured via command line, see available options below, or by running deslant_img -h

Usage

Command line options of deslant_img:

usage: deslant_img [-h] [--data DATA] [--optim_algo {grid,powell}]
                   [--lower_bound LO] [--upper_bound HI]
                   [--num_steps STEPS] [--bg_color BG]

optional arguments:
  -h, --help            show this help message and exit
  --data DATA           directory containing the (.png|.jpg|.bmp) input images
  --optim_algo {grid,powell}
                        either do grid search, or apply Powell's derivative-
                        free optimizer
  --lower_bound LO      lower bound of shear values
  --upper_bound HI      upper bound of shear values
  --num_steps STEPS     if grid search is used, this argument defines the
                        number if grid points
  --bg_color BG         color to fill the gaps of the sheared image that is
                        returned

API

• Import the deslant_img function as shown in the code snippet
• For documentation of parameters see command line parameters above or use help(deslant_img)

from deslant_img import deslant_img
import cv2
import matplotlib.pyplot as plt

img = cv2.imread('data/test1.png', cv2.IMREAD_GRAYSCALE)
res = deslant_img(img)

plt.imshow(res.img)
plt.show()

Algorithm

Vinciarelli and Luettin describe the algorithm in their 2001 paper. Here is a short outline of the algorithm: