Linear Feature Separation From Topographic Maps Using Energy Density and The Shear Transform

Post on 21-Jul-2015

124 views 1 download

Transcript of Linear Feature Separation From Topographic Maps Using Energy Density and The Shear Transform

Presented by

ABHIRAM.S

ROLL NO:01

MTECH COMM ENGG

Guided by

ANOOP CHANDRAN.P

ASST.PROF.

CAARMEL ENG. COLLEGE

1

Introduction

Digitalization of topographic maps is an important data

source of constructing GIS

Maps consist of linear features(roads and condor

lines) and backgrounds(green fields and water bodies)

Linear features are fundamental to GIS ;so its

separation is important

Manual separation is time consuming

Automated separation is based on the colours

2

Contd..

3

When linear feature colour and background colour

are similar then it is difficult to separate them

This paper present a method based on energy density

and shear transform

Shear transform preserves lines directional info

during one directional separation method

Horizontal and vertical templates are used to

separate lines from background

Contd..

Remaining grid background can be removed by grid

template matching

Isolated patches of one pixel and less than ten pixels are

also removed

Union operation on these sheared images give the final

result

4

Existing systems

In 1994 N.Ebi developed a system by converting RGB

colour to another colour space

In 1994 H.Yan proposed a system based on fuzzy theory

;which combines fuzzy clustering and neural n/w’s

In 1996 C.Feng developed a system based on colour

clustering

In 2003 L.Zheng developed a system of fuzzy clustering

based on 2D histogram

5

Contd.. In 2008 Aria Pezeshek introduced a semi automated

method; in this method contour lines are removed by an

algorithm based intensity quantization followed by

contrast limited adaptive histogram equalization.

In 2010 S.Leyk introduced a segmentation method which

uses information from local image plane, frequency

domain and colour space

All methods described above work where the colour

difference b/w line and background is seperable

6

Characteristic Analysis of Linear

Features and Background

Colour based separation is difficult in some case

7

Figures show histogram of image in lab colour space

The are number of peaks in the histogram of first image

But in second image; colour of pixels are close to each

other ; so there is only one peak in the histogram.

It is very hard to separate the line from background of

second image.

8

This figure shows a binary image with complicated

background

Some portion of the image is ideal and other is

complicated

9

1

Ideal portion of background can be removed by using the

Grid templates shown

If the centre pixel and adjacent 8 pixels satisfy the fig 4(a)

and 4(a1) then the pixel is treated as background and

replaced by 1/white

If the centre pixel and adjacent 8 pixels satisfy the fig 4(b)

and 4(b1) then the pixel is treated as line info and

replaced by 0/black

In the fig 3(c) it is a portion of image with complicated

background; it cannot be operated with our grid template

matching

10

Energy characteristics Energy of an image is given by

i=1,2,3...Mj=1,2,3...N

M and N are the height and width of image f(i,j) is the gray value of pixel

Energy of one pixel f(i,j)i-k<m<i+kj-k<n<j+k

size of window w=2k+1

11

Line in gray s/m is dark; but HVS is more sensitive to

brightness so we take negative of gray image

The figure shows that, the energy of negative image is

concentrated on lines

Here the distribution of line and background in ideal case

is shown here

12

The figure shows the distribution of line and background

in the case of actual image

Here fig c. Represents the background and fig d.

Represents the line

13

The histogram of line in fig b. is shown in fig d.

It has only few pixels but all of its energy concentrated on

the lines

Energy ranging from 2.5*104 -3* 104 ;extreme case it is

6* 104

Energy of background is also in the same range; but

energy conc. is higher for lines

14

15

Horizontal and vertical templates are used to separate

lines from background

h2 corresponds to line, it is selected adaptively by

experience; generally 2*2

h1 and h3 corresponds to background pixels generally of

size 4*2 and 2*4

Energy density of the template is

2

Edk = /m*n

m*n-area of template

k=1,2,3

Edk =energy density of each area of template

16

Proposed method

Traditional colour based system fails when the colour of

background and the colour of the lines are similar

This method is based on energy density

Energy density of a negative image is defined as the

average energy in an area

2

Ed = /M*N

M*N-size of area

Ed =energy density

17

Rules for line separation Rule 1:

energy of line is distributed in small area so

energy density is high

energy of background is distributed in large area

so energy density is low

Ed2>Ed1

Ed2>Ed3

ie: energy density of line >energy density of background

18

Rule 2:

if line and background cannot be separated by rule 1

, it is necessary to control the energy difference of line

and background to a certain range

Ed’=Ed1+Ed2+Ed3/3

T=Ed2-Ed’+α

α is acquired by experience; α =3000-5000

Ed2-Ed1>T

Ed2-Ed3>T

h2 is treated as line if and only if Ed2 satisfies rule 1 and

rule 2

19

Background pixel h1 and h3 and isolated patches of one

pixel or less than ten pixels are removed

Finally union operation is performed on the two images

20

Shear transform

Shear transform is a linear transform that displaces point

in a fixed direction

Introduced to avoid the separation difficulties while

operating lines with many direction

Ws,k is the shear operation

s=0,1 k€[-2ndir ,2ndir]

f’s,k(x,y)=f(x,y)*Ws,k

Total number of sheared image is given by

2ndir+1+1

21

Shear transform is performed by sampling pixel according

to the shear matrix

S=0 operation is performed in horizontal direction

S=1 operation is performed in vertical direction

(x’,y’)=(x,y)S1=(x,y)

22

This is the result of shear transform of s=0, ndir=2

so a total of 9 images; union of these images gives a

perfect map

23

Steps of proposed method

STEP 1:

colour image is converted into gray image

Gray=0.233R+0.587G+0.114B

negative of the gray image is taken

I=e*255-gray

‘e’ is a matrix with same size of gray matrix with

all elements equal to one

STEP 2:

Apply shear transform to the negative image

24

Contd.. STEP 3:

Establishment of template: horizontal and

vertical

STEP 4:

Linear feature separation from background:

i.e. : energy of each area in template is calculated,

line is separated from background by rule 1 and

rule 2 with α =4000

25

Contd...

STEP 5:

Removal of miscellaneous point:

i.e. : remaining grid background can be removed

by grid template matching and isolated points can

also be removed

STEP 6:

Inverse shear transform and union operation

26

27

Experiments and Discussions

This is a 342*198 size 7 colour topographical map image

Colour of linear feature and background are similar here

so it is very difficult to separate lines from background

28

Here size of h2 is 2*2

h1&h3 is 4*2 if vertical template is used

h1&h3 is 2*4 if horizontal template is used

α=4000

Fig(b) is the gray image

Fig(c) is the negative image

29

The first set of figures shows the sheared images with k=-1, k=0, k=1

Second set shows energy density based extraction bytemplates

30

Fig (a) shows the union operation of a2, b2, c2,

Fig (b) shows lines with colour info extracted from colour

image

Fig (c) shows the remaining background

31

Comparison of different methods

32

Conclusion

This paper proposes a method to linear separation from

background

Here shear transform is used to overcome the limitation

of directions for lines

Energy density concept is introduced to separate lines

from background

The new method can easily be applied to maps for

efficient separation of lines

Adaptive size fixing of template is a draw back of this

method

33

Reference R. Samet and E. Hancer, “A new approach to the

reconstruction of contour lines extracted from topographic maps,” J. Visual Commun.

E. Hancer and R. Samet, “Advanced contour reconnection in scanned topographic maps,”

H. Chen, X.-A. Tang, C.-H. Wang, and Z. Gan, “Object oriented segmentation of scanned topographical maps,”

S. Leyk, “Segmentation of colour layers in historical maps based on hierarchical colour sampling,” in Graphics Recognition. Achievements, Challenges, and Evolution (Lecture Notes in Computer Science),

34

Thank you

35

Questions?

36