Hough Transform Algorithm for Biology-image Recognition ... · An artificial retina camera (ARC) is employed for real-time preprocessing of the input image. And the algorithm of Hough

1

Hough Transform Algorithm for Biology-image Recognition Using an Artificial Retina Camera

Xin Lin and Kazunori OtobeNational Agriculture Research Center, Japan

Xin Lin National Agriculture Research Center

Annual Meeting 2000

Process for Pattern Recognition

● Observation CCD image sensor

● Preprocessing with the help

● Character detection of computer

● Recognition/Deciding algorithm

Hough Transform for Object Detection

● Good effect for detecting

and finding a noisy image

● For a curved object Hough space

will become high dimensionality

Hough Transform Algorithm for Biology-image Recognition Using an Artificial Retina Camera

2

● AbstractAn artificial retina camera (ARC) is employed for real-time preprocessing of the

input image. And the algorithm of Hough transform is advanced for detecting a

circle-edge object in the two-dimension space. This method also works in parallel

for processing multiple input patterns and partial input images.

● Image processing using an ARC

● Image detecting by the Hough transform

● Experiments of pattern recognition

● Results and discussion


Annual Meeting 2000Hough Transform Algorithm for Biology-image Recognition Using an Artificial Retina Camera

3

● Principles of image processing using ARC


The structure of ARC. (256 gray level, 128×128 pixels) Computer

AR LSI( VSPD, scanner,

multiplexer )

MCU( Image recognition

algorithm )

Image Data


(b)(a) (c)

Examples of image processing using the ARC. (a)original, (b) edge enhancement,and (c) binarization.

The ARC is acombination of an AR-LSI and a MCU. TheAR-SLI includes avariable sensitivityphotodetection cell(VSPD), a scanner, and amultiplexer for paralleloptics computation andsignals output. TheMCU is a micro-controlunit to control the AR-LSI and recognition theoutput signal from theAR-LSI.

4Xin Lin National Agriculture Research Center

0 2 4 6 8 10

0

1

2

3

4

5

6

θ0

ρ0

( xi, yi )

ρ0 = xcosθ0+ysinθ0

y

ximage space (x-y plane)

0.0 0.5 1.0 1.5 2.0 2.5 3.0

-5

0

5( ρ0, θ0 )

loci: ρ = xicosθ+yisinθ

θ

ρ

parameter space (ρ−θ plane)

● Hough transform for detecting ellipses (include circles)

( ) ( ) ρφφθφθ =++=−+− cossinsinsincos ,1// 220

220 zyxbyyaxx


● The principle of the Hough transformation for pattern detection● Hough transform for detecting straight lines

The Hough techniquetransforms the x-y planeinto the ρ -θ plane.

(1) an x-y point becomes asin curve in ρ -θ plane;

(2) a ρ -θ point becomes astraight line in x-y plane.

If there is a peak inparameter space and it isthe cumulative value ofall the sin curves, then astraight-line can bedefined in x-y plane.

The Hough technique forrecognition of ellipses orcircles is quite complexbecause of the Houghspace become a highdimensionality.


● HT-I space

),,,( 1 θααρ nH Κ= parameters of the curve:,,1 nαα Κ

The center of thecircle at the origin

rrHH cn === ),(),,,( 1 θθααρ Κ

Circular objectis not centered ⎩

⎨⎧

<−++−+−≥−+−+

=0)cos( )cos(

0)cos( )cos(απθαθ

αθαθρ

trtrtrtr

)(tan , 0012

02

0 xyyxt −=+= α

● HT-Ⅱspace

rrH c μμ ρθρρ ==′ ),(

x0, y0 : origin location

Hough Transform Algorithm for Biology-image Recognition Using an Artificial Retina Camera Annual Meeting 2000

● An algorithm for detection circles using straight-line Hough transform

•If the input image is acontinuous curve, then thispattern in the Hough space willalso be a continuous curve;

•Using a transformation to mapthe HT-I space to a new HT-IIspace;

•Performing an inverse HT onHT-II space, we will obtain apeak at (x0, y0 );

•When the radius of the circle isnot known, With the help oftrial-and-error, we can alsoobtain this radius value.

(1) forming the straight-line HT;

(2) transforming HT-I space;

(3) performing an inverse HT.


● Experiment I ―― input images (a cabbage)

(a) original (b) binarization threshold (c) edge extraction


● A cabbage is used for a input image. The size is 128×96 pixels.

●We can see the edge information is parts of two circles (internal circle and external circle).


0.0 1.5 3.0 4.5 6.00

5

10

15

20

25

30

35

external circle

internal circle

HT-I space

ρ

θ

● Experiment I ―― Results of HT-I and HT-II

(a) The patterns which consist of thepeaks in HT-I space after thresholding.They are parts of the cos curves.

(b) The mapping patterns, whichcorresponds to the radius of twocircles in HT-II space


0.0 1.5 3.0 4.5 6.0-2.0

-1.5

-1.0

-0.5

0.0

0.5

1.0

1.5

2.0

external circle

internal circle

HT-II space

ρ

θ


rS = 24 pixels rL = 30 pixels

The circle rS centered at (1, 0) and rL is (1, -1) pixels

Extractive circles

● Experiment I―― inverse HT and circles detection

(c) The result of inverse HT. Thecoordinates of the cumulative pointsdenote the origin positions of two circles.

Hough Transform Algorithm for Biology-image Recognition Using an Artificial Retina Camera Annual Meeting 2000

0.0 1.5 3.0 4.5 6.00

1

2

external center

internal center

inverse HT space

t

α

(d) Extractive circles. Their radiuses are24 and 30 pixels, and centers at (1, 0)and (1, -1), respectively.


(a) original (b) binarization

(c) edge


● Experiment II ―― input images (multiple cabbages)

● There are multiple circles in theinput image.

● The size of input image is128×80 pixels.


0.0 1.5 3.0 4.5 6.00

10

20

30

40

50

60

HT-I space

circle 4 circle 5

circle 7circle 6circle 3 circle 2

ρ

θ


● Experiment II―― Results of HT-I and HT-II

(a) The patterns in HT-I space afterthresholding. They are parts of thecos curves.

(b) The mapping patterns, whichcorresponds to the radius of sixcircles in HT-II space.

0.0 1.5 3.0 4.5 6.0-60

-40

-20

0

20

40

60HT-II space

ρ

θ

circle 5

circle 7

circle 6

circle 4

circle 3

circle 2


r2,3,4,5,7 = 13 pixels r6 = 14 pixels

r1 failing, r2 (-13, 13), r3 (17, 13), r4 (43, 13), r5 (-41, -14), r6 (-14, -14), r7 (13, -14), r8 failing.

Extractive circles

1 2 3 4

5 6 7 8


0 1 2 3 4 5 60

10

20

30

40

50

60

inverse HT space

circle 7circle 6

circle 5circle 4

circle 3circle 2

t

α

● Experiment II ―― inverse HT and circles detection

(c) The origin positions of six circlesfrom inverse HT.

(d) Radius and origins of extractive circles.The cabbages 1 and 8 cannot be detectedbecause their edges are not a circle.


10 11 12 13 14 1545

50

55

60

65

70

Est

imat

ion

diam

eter

(pi

xel)

Measurement diameter (cm)

● Experimental result

We have also detected four cabbages and their radius is different.This is the result for the relation between the measurement andestimation values by our algorithm. We can see there is a good

linear is showed.


● Conclusion

●We have described a method forpattern recognition.

● An ARC is employed for real-time preprocessing of the inputimage.

● The algorithm of the straight-

line HT is advanced for detectingcircular object in the two-dimension space.

● This method also works inparallel for multiple inputs andfor partial input curves.

Hough Transform Algorithm for Biology-image Recognition ... · An artificial retina camera (ARC) is employed for real-time preprocessing of the input image. And the algorithm of Hough

Documents