Introduction

Text Detection and Recognition

Problem Domain and Assumptions

Given an RGB image , I have to give the text present in
it.
Text detected can be either on a dark background or a
dark text on a light background.

Stroke Width Transform

It can be broken down into 3 steps: SWT, grouping
pixels into letter candidates based on their stroke width
and finally grouping letter candidates into regions of
text.
A stroke in the image is a continuous band of nearly
constant width.

SWT

Flowchart

Finding letter candidates

This is done by first grouping pixels with similar stroke
width and then applying several rules to distinguish the
letter candidates like variance of stroke width within a
component must not be too big, aspect ration of a
component must not be within a small range, ignore
components whose size is too large or too small.

Grouping letter candidates into text

regions
This is mainly removing false letter candidates.
Some set of rules like two letter candidates should have
similar stroke width, ratio between height and width of
letters should be less than 2.5, distance between letters
must not exceed three times the width of wider one.

Results

Results

Recognition
Divided into 3 parts:
1) Segmentation
2) Feature Vector
3) Training
Coding for Segmentation and feature vector done.
Features taken:
4) Histogram Projection based on mean distance
5) Histogram Projection based on pixel value
6) Vertical zero crossing
7) Horizontal Zero crossing