Você está na página 1de 4

IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 21, NO.

6, MAY 2012

2955

i An Edge Detection Improved Algorithm Based on Morphology and Wavelet Transform


XIA Kai-jian, CHANG Jin-yi , YAO Yu-feng , ZHONG Shan

AbstractAn improved edge detecting algorithm based on mathematical morphology and wavelet transform is proposed to overcome the limitation which embarrasses the performance of the traditional mathematical morphological methods. In the wavelet domain, the low-frequency sub-image edges are detected by solving the maximum points of local wavelet coefficient model to restore edges, while the high-frequency sub-image edges are detected by multi-scales and two-structuring elements mathematical morphology. Finally it can get a complete edge of the image. Experimental results showed that compared with the traditional wavelet transform edge detecting method and math morphology method, this method can adaptively extract accurate. edge information, and better decrease the noise. It is an effective edge detection method. Keywords- wavelet transform; math morphology; edge detection; multi-structuring elements; noise I. INTRODUCTION In image processing, the edge detection is an important image pre-processing process, and it locates an important position in the field of computer vision and image processing. It is widely used in target recognition, image segmentation, remote sensing, medical image analysis and other fields. The edge of the image expressed image signal of the mutation, which contains a lot of information that can be used for image analysis, object segmentation, image filtering and image recognition. Many of the traditional edge detection methods, such as the common types of algorithms, Log operator, Sobel operator, Canny operator[1] and the gradient operator and so on, but a large number of digital image processing results show that these pairs of directional edge detection

operator more sensitive, anti-noise ability is poor, generally difficult to obtain satisfactory test results. The edge detection method based on wavelet transform [2-4] has a good localization properties, it is a good tool for edge detection using wavelet transform multi-scale analysis capabilities for the image to provide a coarse-to-fine multi-resolution representation, and the application of these characteristics can be achieved right image edge extraction. Mathematical morphology [5] is a nonlinear filtering method; applied mathematical morphology can be simplified image data to maintain their basic shape characteristics. Morphology in the continuity of image edge detection and various directional aspects superior to other traditional methods [6-8], due to a single structural element can only detect structural elements of the edge information in the same direction, while the other direction is not sensitive to the edge. This paper selected multi-scale and twostructuring structure elements morphology edge detection operator. The selection of diamondshaped and cruciform logical combination of two kinds of structural elements that can detect edges in different directions and better ensure the integrity of edge information. Combining the advantages of the two methods, the method that the combining of the math morphological method and wavelet transform edge detection was extended. Using this method made the edge information more accurate and stronger anti-noise performance. Experimental results show the feasibility and superiority of the algorithm. II. MULTI-SCALE EDGE DETECTION BASED ON WAVELET TRANSFORM In the following, two-dimensional wavelet transform, (x, y) said the Gaussian smoothing function [9], so that

978-1-4244-5586-7/10/$26.00 2012 IEEE

2956

IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 21, NO. 6, JUNE 2012

( x, y) 1 2 ( , ) s s s 1 1 x y 1 (x, y) ( , )
s

x y

1 2 3

s s s2 2 2 x y 1 s (x, y) 2 ( , ) s s s For f (x, y) convolution processing,


1 Ws

F B(x, y) max{F(x s, y t) B(s, t)} 9 Similarly, the opening operation and closing operation are mutually dual operations. The opening operation is mainly to delete relatively small objects, and remove some unwanted content of the edge, while the closing operation has the role of filling within a small hollow. Mathematical definition is as follows: 3. Opening operation: F o B (FB) B
10 11 4. Closing operation:

f (x, y) f

* 1

5 f (x, y) f (x, y) 1 2 Where, Ws f (x, y) and Ws f (x, y) constitute the two-component of the scale s two-dimensional wavelet transform. For the scale s, the model of the gradient vector:

2 Ws

s *

(x, y)
2 s

F B (F B)B

Ms f (x, y) | Ws1 f (x, y) |2 | Ws2 f (x, y) |2


Its delta amplitude:
As

f ( x, y) arctan(

W s ( x, y)
s
W 1f ( x, y)

The gradient vector direction is the direction of gradient local modulus maxima, so we can get the edge point of image by detecting the local modulus maxima of wavelet transform coefficients along the gradient vector direction. In this paper, it uses wavelet coefficients using the local modulus maxima to restore the image edges, and avoids the adverse effects of the established image edge because of artificial thresholds, thus significantly improve the generalization ability of the algorithm. III.EDGE DETECTION BASED ON MATHEMATICAL
MORPHOLOGY

B. Multi-scale and two-structuring structure elements morphology edge detection operator In the edge detection based on the morphology studies, if you want to make morphological edge detection operator has the function of removing image noise, then the structuring element of the morphological edge detection operator must be greater than or equal to the noise point scale. Because large-scale structural elements has greater function that remove the image noise, and it is beneficial to determine the broad contours of original image, but some of the smaller details of the edge will be removed as a noise in the original image. If using a smaller structuring element, it is still difficult to eliminate the image noise points, but smallscale structuring element can better maintain a edge detail. In this paper, morphological gradient operator has been improved, that is, it extended a compromise method, which uses multi-scale and two-structuring structure elements for alternating sequence of morphological opening and closing filtering in order to smooth image and remove the image noise. It effectively solves the problems posed by single structural elements. The Multi-scale and two-structuring structure elements morphology edge detection operator:

E(F ) (F o B) A (F B)A

A. Morphological basic operations Mathematical morphology (MM) is a theory and technique for the analysis and processing of geometrical structures, based on set theory, lattice theory, topology, and random functions. Its basic idea is to use a certain form of structural elements to measure and extract the corresponding shape of image in order to achieve the right image analysis and recognition purposes. The application of mathematical morphology can be simplified image data to maintain their basic shape features, and to remove extraneous structures. There are four basic operations in the mathematical morphology: expansion, corrosion, open and closed. Where, the operations of expansion and corrosion are Dual operation with each other, and the corrosion is mainly to reduce the process of an object along the border, while the expansion process of the merger. Where F (x, y) is gray image, B(s,t) is Structuring element. Mathematical definition [10] is as follows: 1. Corrosion operation

13 Where A and B is respectively diamond-type 5 5 structuring element and cross-shaped 3 3 structuring element, Because A and B are structural elements with different scale, for the smallscale structure element B, although the ability that removes noise is weak, it can better maintain the image edge information. And for large-scale structural elements A, it can well remove the image noise, but it will lose some edge information. So Multiscale and two-structuring structure elements that I proposed not only can

0 0 1

0 0 1

A= 0 1 1 1
1 1 1 0 1

0 1 0 B= 1 1 1 0 1 0

1 1 1 0 1 0 0

FB(x, y) min{F (x s, y t) B(s, t)}


2. Expansion operation

8
B 1

0 0

LAW et al.: An Edge Detection Improved Algorithm Based on Morphology and Wavelet Transform

2957

effectively eliminate the image noise, but also can effectively IV.EDGE DETECTION ALGORITHM BASED ON MORPHOLOGY maintain good edge information. AND WAVELET TRANSFORM Firstly, the image was decomposed based on the wavelet in this paper can be more effectively filter out the noise, transform algorithm, then for the high-frequency and edge details are retained. coefficients of the decomposed sub-image, uses mathematical morphology approach based on multi-scale and two-structuring structure elements for the edge detection. Concrete steps are as follows: Step1. The first-order partial derivative of Gaussian function was used as the basic wavelet. And the image was decomposed for N times. Step2. For the low-frequency coefficients of the decomposed sub-image, uses modulus maxima for the edge (a) Original image (b) Canny operator detection Step3. For the high-frequency coefficients of the decomposed sub-image, uses mathematical morphology approach based on multi-scale and two-structuring structure elements for the edge detection. We used formula (13) morphological operators to detect its edge of sub-image. Step4. Use all the coefficients of the above-mentioned processing to reconstruct image. Step5. Smooth the image, eliminate the edge of false (c) Wavelet algorithm (d) morphology information, connect gradually broken edge points, and obtain a single, continuous edge. V. EXPERIMENTAL RESULTS AND ANALYSIS In the Matlab6.5 experimental environment, according to the above proposed methods for image edge detection test, this algorithm is simple and easy to implement. In order to test the effectiveness of the algorithm, diamond-type 5 5 structuring element and cross-shaped 3 3 structuring element were selected, and Gaussian function was used as the basic wavelet. Taking into account the speed of calculation and the requirement of accuracy, we made wavelet decomposition for two times, that is N=2. The experimental results shown in fig (1) and fig (2), fig (1) show an image without Gaussian function, where, fig (1a) shows original image, fig (1b) shows the edge detection result for Canny operator, Fig (1c) shows edge detection result for Wavelet algorithm, Fig (1d) shows edge detection result for mathematical morphology, and experimental results for the proposed method are shown in Fig (1e). Fig (2) show an image with Gaussian function, from the experimental results, we can see that the algorithm proposed

(e) Our proposed method


Fig (1) Comparison chart of different algorithms without image noise

(a) Original Lena with noise

(b) canny

(c) (c)wavelet
(d) morphology

(e) Our proposed method

IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 21, NO. 6, JUNE 2012

2958

Fig (2) Comparison chart of different algorithms with image noise

VI.CONCLUSION In this paper, an improved edge detecting algorithm based on mathematical morphology and wavelet transform is proposed. For the image with noise, firstly, we decomposed it based on the wavelet transform algorithm. In the wavelet domain, the low-frequency sub-image edges are detected by solving the maximum point s of local wavelet coefficient model to restore edges, while the high-frequency sub-image edges are detected by multi-scales and two-structuring elements mathematical morphology, So that we can get a relatively good image characteristics and details. The experimental results show the feasibility and effectiveness of the algorithm. Compared with the other edge detection algorithm, our algorithm not only has a stronger anti-noise, but also better able to maintain the integrity of the edge information. ACKNOWLEDGMENT
The 1 author would like to thank Professor Chang jinyi for helpful discussions and proofreading this paper.
st

REFERENCES
[1]

[2] Sun J X ,Gu D B. A multiscale edge detection algorithm based on wavelet domain vector hidden Markov tree model [J].PatternRecognition,2004,37(7):1315-1324 [3] Zhang L, Bao P. Edge detection by scale multiplication in wavelet domain [J]. Pattern Recognition Letters. 2002.23(14):1771 1784 [4] Guo L Z,Zhao J H. Edge Detection Based on Wavelet Transform [J]. Journal of Qingdao University of Technology 2007 28 2 7880 [5] Tan C Q. Mathematical Morphology Method and Application [M].Beijing: Science and Technology Press,1990 [6] Fu Y Q, Wang Y S. An algorithm for edge detection of grayscale image basedon mathematical morphology [J]. Journal of Harbin Engineering University, 2005, 26(5):685-687.

Canny J E. A computional approach to edge detection[J].IEEE Trans on PAM 1,1985.8(6):679-698

[7] Yang H, Zhang J W. Research on Application of Mathematical Morphology in Edge Detection of Image [J]. Journal of Liaoning University,2005,32(1) 50-53 [8] Liu Q, Lin S S. Edge detection method based on math morphological transforms [J]. Journal of South China University of Technology 2008 36 9 :113-116 [9] Lin S Y Su G C, Chen D, Han X G. Edge Detection Based on Wavelet Transform and Morphology [J]. Chinese Journal of Scientific Instrument 2004 25 4 685-687 [10] Cui Y. Image Processing and AnalysisMathematical Morphology Method and Application [M].Beijing Science and Technology Press 2000

Você também pode gostar