Arabic handwriting recognition system and method

Patent metadata

Field	Value
Publication number	US-10163019-B1
Application number	US-201816035257-A
Country	US
Kind code	B1
Filing date	Jul 13, 2018
Priority date	Sep 19, 2017
Publication date	Dec 25, 2018
Grant date	Dec 25, 2018

Abstract

A system, a non-transitory computer readable medium, and a method for Arabic handwriting recognition are provided. The method includes acquiring an input image representative of a handwritten Arabic text from a user, partitioning the input image into a plurality of regions, determining a bag of features representation for each region of the plurality of regions, modeling each region independently by multi stream discrete Hidden Markov Model (HMM), and identifying a text based on the HMM models.

First claim

The invention claimed is:

1. Arabic handwriting recognition method, comprising:
acquiring, an input image from a document representative of a handwritten Arabic text from a user, wherein the input image is acquired with at least one selected from the group consisting of a scanner, a camera and a data storage device;
partitioning, using processing circuitry of a server, the input image into a plurality of regions;
determining, using the processing circuitry, a bag of features representation for each region of the plurality of regions, wherein the bag of features representation includes one or more mid-level features selected from the group consisting of a percentile of intensities, an angle, a correlation, and an energy;
modeling, using the processing circuitry, each region independently by multi stream discrete Hidden Markov Model (HMM); and
identifying, using processing circuitry, a recognized text based on the HMM models.

2. The method of claim 1 , further comprising:
smoothing the input image by a Gaussian Kernel proportional to a descriptor spatial area.

3. The method of claim 1 , wherein the plurality of regions includes a middle region, an upper region, and a lower region.

4. The method of claim 3 , wherein the middle region includes a writing baseline.

5. The method of claim 1 , further comprising:
applying a sliding window for each of the plurality of regions;
partitioning each window of the sliding window into one or more cells; and
determining one or more local descriptors for each cell.

6. The method of claim 5 , wherein a center associated with the one or more local descriptors matches a center of each window.

Assignees

• Univ King Fahd Pet & Minerals

Inventors

• Mahmoud Sabri A
• Assayony Mohammed O

Classifications

• G06F40/174

  Form filling; Merging · CPC title

• G06V10/464

  using a plurality of salient features, e.g. bag-of-words [BoW] representations · CPC title

• G06F3/018

  Input/output arrangements for oriental characters · CPC title

• G06T7/11

  Region-based segmentation · CPC title

• G06K9/4676

  Physics · mapped topic