Chinese Character Recognition for Video

1
Chinese Character Recognition for Video

• Presented by Vincent Cheung
• Date 25 October 1999

2
Introduction

• Many dialects in Chinese, but Chinese Characters
  is common in anywhere.
• Many video programs have Chinese subtitles
  nowadays
• Extract text from digital video programs can help
  for indexing, searching and retrieval

3
Features of Subtitles

• Characters are in foreground
• They are monochrome
• They are rigid, from frame to frame
• They are upright
• They have size restrictions
• They contrast with the background
• They appear in clusters at a limited distance
  aligned to a horizontal line

4
Steps to Recognise Text

• Clearing the background, removing noise
• Segmenting the characters
• Recognising them by pattern matching

5
Demo Video

• A piece of news from ATV about Airport Authority
  Hong Kong and is reported in Cantonese
• In MPEG format
• 1... 2... 3... Action!

6
MPEG Video

• Consisted of a video track and an audio track
• Consisted of frames
• For video part, a frame is representing a static
  image

7
Steps to Remove Background

• Agnihotri Dimitrova Suggested 7 steps
  procedures
• Channel Separation
• Image Enhancement
• Edge Detection
• Edge Filtering
• Character Detection
• Text Box Detection
• Text Line Detection Enhancement

8
Sample Frame

• The 100th frame of the demo video

9
Channel Separation

• Use Red Channel which gives higher contrast edges
• More probably that natural environment are in
  blue or green

Green Channel
Blue Channel
Red Channel
10
Image Enhancement

• To filter salt and pepper noise
• To sharpen the edges
• Quality of our mpeg video is quite good that we
  no need to take this step

11
Edge Detection

• Find out the edges from the image
• Use a 3x3 matrice mask
• -1 -1 -1
• -1 12 -1
• -1 -1 -1
• Use Sobel Filter instead
• edges around text may be broken and not connected

12
Sample Edge Image
13
Edge Filtering

• To remove areas which possibly do not contain
  text
• Characters would give high density of objects,
  hence high density of edges
• Finding out areas with high density of edges
  which give hints of where the characters located

14
Density of edges in horizontal lines
15
Filtering the Irrelevant Edges
16
Density of Edges in Vertical
17
What if the length of subtitle is short??

• Cut the image into certain parts and calculate
  the density of edges in those areas
• Prevent the case if the subtitle is short and
  cannot give an overall view

18
Sample Image Divided in Parts
19
Challenges in Chinese Characters Segmentation

• Square?
• Not Really, they are variable in size!! Having
  different height and width
• e.g. (?, ?)
• Lead to some problem in Fixed-Distance Approach
  Segmentation
• More problems if mixed with English, Numbers, and
  Symbols
• e.g. 18??IBM???

20
Challenges in Chinese Characters Segmentation

• Usually written in horizontal way, like English.
• Do segmentation like English?
• English each character is horizontally linked
• Chinese may not have such linkage
• e.g.?, ?

21
Character Recognition

• Pattern Matching
• most straight forward
• two pattern are compared
• by using pattern distance

22
Classification for Faster Matching

• By blackness (e.g. ?, ?)
• By projection profiles

23
Possible Enhancement

• Picking out the moving objects by keeping track
  of a number of consecutive frames
• Use of lexicon to choose the most possible
  character

24
Q A