Study Image Edges ABSTRACT

and Edge

Comparison Detection

of

Various

Techniques and are

characterize a

boundaries of

therefore

problem

fundamental

importance in image processing. Image Edge detection significantly reduces the amount of data and filters out useless information, while preserving the important structural properties in an image. Since edge detection is in the forefront detection, of it image is processing to for a object good crucial have

understanding of edge detection algorithms. In this paper the comparative analysis of various Image Edge Detection techniques is presented. he software is developed using !" #"$ %.&. It has been shown that the 'anny(s edge detection algorithm performs better than all

these operators under almost all scenarios. Evaluation of the images showed that under noisy conditions 'anny, #o)* #aplacian of )aussian+, ,obert, -rewitt, Sobel e.hibit better performance, respectively. /. It has been observed that 'anny(s edge detection algorithm is computationally more e.pensive compared to #o)* #aplacian of )aussian+, Sobel, Image 1. Edge an detection he refers to the -rewitt Edge and ,obert(s operator. Digital -rocessing I TR!D"CTI! process are of 0eywords1 Detection, 2oise,

identifying and locating sharp discontinuities in image. discontinuities abrupt changes in pi.el intensity which characterize boundaries of objects in a scene. 'lassical methods of edge detection involve convolving

the image with an operator *a 34D filter+, which is constructed to be sensitive to large gradients in the image while returning values of zero in uniform regions. operators available, each here are an to be e.tremely large number of edge detection designed sensitive to certain types of edges. 5ariables involved in the selection of an edge detection operator geometry include and the of Edge Edge operator orientation, structure. determines 2oise he a environment

characteristic direction in which it is most sensitive to edges. 6perators can be optimized to loo7 for horizontal, vertical, or diagonal edges. Edge detection is difficult in noisy images, since both the noise and the edges contain high4 frequency content. "ttempts to reduce the noise result in blurred and distorted

edges. 6perators used on noisy images are typically larger in scope, so they can average enough data to discount localized noisy pi.els. his results in less accurate localization of the detected edges. 2ot all edges involve a step change in intensity. Effects such as refraction or poor focus can result in objects with boundaries defined by a gradual change in intensity 8/9. he operator needs to be chosen to be responsive to such a gradual change in those cases. So, there are problems of false edge detection, missing true edges, edge localization, high computational time and problems due to noise etc. herefore, the

objective is to do the comparison of various edge detection techniques and analyze the performance of the various techniques in

different here are many ways to

conditions. perform edge

detection. :owever, the majority of different methods may be grouped into two categories #radient $ased Edge Detection% he gradient method detects the edges by loo7ing for the ma.imum and minimum in the first he to derivative $ased method "n #aplacian find edges. of Edge searches has the image. Detection% for the zero one4 &ap'acian

crossings in the second derivative of the image edge dimensional shape of a ramp and calculating the derivative of the image can highlight its location. Suppose we have the following signal, with an edge shown by the jump in intensity below1 If we ta7e the gradient of this signal *which, in

one dimension, is just the first derivative with respect to t+ we get the following1 'learly, the derivative shows a ma.imum located at the center of the edge in the original signal. his method of locating an edge is characteristic of the ;gradient filter< family of edge detection filters and includes the Sobel method. " pi.el location is declared an edge location if the value of the gradient e.ceeds some threshold. "s mentioned before, edges will have higher pi.el intensity values than those surrounding it. So once a threshold is set, you can compare the gradient value to the threshold value and detect an edge whenever the threshold is e.ceeded 839. =urthermore, when the first derivative is at a ma.imum, the second derivative is zero. "s a result, another alternative to finding the location of an edge is

to locate the zeros in the second derivative. his method is 7nown as the #aplacian and the second derivative of the signal is shown below1 In this paper we analyzed and did the visual comparison of the most commonly used )radient and #aplacian based Edge Detection techniques. In section 3 the problem definition is presented. In section > the various edge detection techniques have been studied and analyzed. In section ? the visual comparisons of various edge detection techniques have been done by developing software in !" #"$ %.&. Section @ discusses the advantages and disadvantages of various edge detection techniques. Section A discusses the conclusion reached by analysis and visual comparison of various edge detection techniques developed using !" #"$ %.&

(.

)R!B&E*

DE+I ITI!

here are problems of false edge detection, missing true edges, producing thin or thic7 lines and problems due to noise etc. In this paper we analyzed of the and most did the visual used comparison commonly

)radient and #aplacian based Edge Detection techniques for problems of inaccurate edge detection, missing true edges, producing thin or thic7 lines and problems due to noise etc. he software is developed using !" #"$ %. ,. ,.1 he operator ED#E DETECTI! So$e' consists of a pair TEC- I."ES !perator of >B>

convolution 7ernels as shown in =igure /. 6ne 7ernel is simply the other rotated by C&D. ). )y =I)E,E /1 !as7s used by Sobel 6perator

hese

7ernels

are

designed

to

respond

ma.imally to edges running vertically and horizontally relative to the pi.el grid, one 7ernel for each of the two perpendicular orientations. separate he 7ernels can of be applied gradient separately to the input image, to produce measurements the component in each orientation *call these ). and )y+. hese can then be combined together to find the absolute magnitude of the gradient at each point and the orientation of that gradient 8>9. by1 ypically, computed ) which is F much ). faster G to an appro.imate magnitude is )y compute. using1 he gradient magnitude is given

he angle of orientation of the edge *relative

to the pi.el grid+ giving rise to the spatial gradient H ,.( quic7 each F is arctan*)y cross 34D output spatial Ro$ert/s to point compute, in the given I by1 ).+ operator% gradient the he

he ,oberts 'ross operator performs a simple, measurement on an image. -i.el values at represent estimated absolute magnitude of the spatial gradient of the input image at that point. operator consists of a pair of 3B3 convolution 7ernels as shown in =igure 3. 6ne 7ernel is simply the other rotated by C&D8?9. his is very similar hese grid, to 7ernels one the are Sobel designed for each to of operator. respond the two =I)E,E 31 !as7s used for ,obert operator. ma.imally to edges running at ?@D to the pi.el 7ernel

perpendicular orientations. produce these ). separate and

he 7ernels can be of then the be

applied separately to the input image, to measurements hese to find can the gradient component in each orientation *call )y+. combined together absolute

magnitude of the gradient at each point and the orientation of that gradient. magnitude is ) which is F much is computed ). faster G to given he gradient by1 using1 )y compute.

although typically, an appro.imate magnitude

he angle of orientation of the edge giving rise to the spatial gradient *relative to the pi.el grid H ,., F orientation+ arctan*)y I )re0itt/s is ).+ given J >K I by1 ?

operator%

-rewitt operator 8@9 is similar to the Sobel operator and is used for detecting vertical and horizontal 4/ 4/ 4/ ). detector ,.1 3nd &ap'acian spatial derivative of of an #aussian% image. he he #aplacian is a 34D isotropic measure of the #aplacian of an image highlights regions of rapid intensity change and is therefore often used for edge detection. he #aplacian is often applied to an image that has first been smoothed with something appro.imating a )aussian Smoothing filter in order to reduce its edges & & & in images. G/ G/ G/ )y

=I)E,E >1 !as7s for the -rewitt gradient edge

sensitivity to noise.

he operator normally

ta7es a single gray level image as input and produces another gray level image as output. he #aplacian #*.,y+ of an image with pi.el intensity #*.,y+F L values L .3 I*.,y+ 3I is G L given L by1 3I y

Since the input image is represented as a set of discrete pi.els, we have to find a discrete convolution 7ernel that can appro.imate the second derivatives in the definition of the #aplacian8@9. 7ernels are hree commonly in used =igure small ?. shown

$ecause these 7ernels are appro.imating a second derivative measurement on the image, they are very sensitive to noise. before applying the #aplacian filter. o counter his pre4 this, the image is often )aussian Smoothed

processing step reduces the high frequency noise components prior to the differentiation step. In fact, since the convolution operation is associative, we can convolve the )aussian smoothing filter with the #aplacian filter first of all, and then convolve this hybrid filter with the image to achieve the required result. Doing things this way has two advantages1 Since both the )aussian and the #aplacian 7ernels are usually much smaller than the image, this method usually requires far fewer arithmetic operations. he #o) *M#aplacian of )aussianN+8A9 7ernel can be pre4calculated in advance so only one convolution needs to be performed at run4time on the image.

I TR!D"CTI! Edge detection refers to the process of identifying discontinuities and in locating an image. sharp he

discontinuities are abrupt changes in pi.el intensity which characterize boundaries of objects in a scene. 'lassical methods of edge detection involve convolving the image with an operator *a 34 D filter+, to zero be in which sensitive uniform to is large constructed values of

gradients in the image while returning regions. here are an e.tremely large number of edge detection operators available, each

designed

to

be

sensitive

to

certain types of edges. 5ariables involved in the selection of an edge detection operator include structure. determines Edge he geometry of the operator a characteristic orientation, 2oise environment and Edge

direction in which it is most sensitive to edges. 6perators can be optimized to loo7 for horizontal, vertical, or diagonal edges. Edge detection is difficult in noisy images, since both the noise and the edges result in contain blurred and highfrequency distorted on edges. noisy content. "ttempts to reduce the noise 6perators used

images are typically larger in scope, so they

can

average

enough

data

to

discount pi.els.

localized

noisy

)R!B&E*

DE+I ITI!

here are problems of false edge detection, missing true edges, producing thin or thic7 lines and problems due to noise etc. In this paper we analyzed and did the visual comparison of the most commonly used )radient and #aplacian based Edge Detection techniques for problems of inaccurate edge detection, missing due to true noise edges, etc. producing thin or thic7 lines and problems

Canny

Edge

Detection

A'gorithm

he 'anny edge detection algorithm is 7nown to many 'annyNs as the optimal edge to detector. intentions were

enhance the many edge detectors already out at the time he started his wor7. :e was very achieving his successful goal and his ideas to in and Edge

methods can be found in his paper, O" 'omputational "pproach DetectionO8//9. In his paper, he followed a list of criteria to improve current methods of edge detection. he first

and most obvious is low error rate. It is important that edges occurring in images should not be missed and that he there be no responses to non4edges. well localized. In

second criterion is that the edge points be other words, the distance between the edge pi.els as found by the detector and the actual edge is to be at a minimum. " third criterion is to have only one response to a single edge. implemented completely multiple detector eliminate responses first because the to the possibility an his was the of

first two were not substantial enough to edge. to

$ased on these criteria, the canny edge smoothes image

eliminate

and

noise.

It

then

finds

the image gradient to highlight regions with high spatial derivatives. trac7s at now the further is along ma.imum he used reduced to that he algorithm then these *nonma.imum array along not is the been by trac7 have hysteresis.

regions and suppresses any pi.el that is not suppression+. :ysteresis remaining gradient

pi.els

suppressed. :ysteresis uses two thresholds and if the magnitude is below the first threshold, *made it a is set non to zero edge+.

C! C&"SI! S Since edge detection is the initial step in object recognition, it is important to 7now the differences between edge detection techniques. In this paper we studied the most commonly used edge detection Detection. !" #"$ )radient4based being very algorithms sensitive such to as techniques of )radient4based and #aplacian based Edge he software is developed using %.&. the

-rewitt filter have a major drawbac7 of noise. he size of the 7ernel filter and coefficients are fi.ed and cannot be adapted to a given image. "n adaptive edge4detection algorithm is necessary to

provide a robust solution that is adaptable to the varying noise levels of these images to help distinguish valid image contents from visual artifacts introduced depends heavily , by on the which noise. adjustable is the he performance of the 'anny algorithm parameters,

standard deviation for the )aussian filter, and the threshold values, P /( and P 3(. also controls the size of the )aussian filter. becomes. he bigger the value for , his implies more

the larger the size of the )aussian filter blurring, necessary for noisy images, as well as detecting larger edges. "s e.pected, however, the larger the

scale of the )aussian, the less accurate is the localization of the edge.

,eference1 http1IIseminarprojects.comI hr ead4study4and4comparison4of4various4 image4edge4detection4techniques4full4 reportQi.zz3 pI2ER'0