[attachment=1160]

---

ABSTRACT

Edge detection is one of the most important steps leading to the analysis of processed image data. Image texture segmentation is an important problem and occurs frequently in many image-processing applications. The image can be segmented by detecting there boundaries. This paper presents some algorithms that are required for Edge detection of textured image. Algorithms for Edge detection are presented that works directly for non-textured image. These algorithm can be carried out for edge detection of textured images also by preprocessing the image through discrete Filters. Experimental results on images containing various synthetic and natural textures have been carried out and a comparison of existing techniques is shown.
1. INTRODUCTION
Edge detection is the process of detecting the edges of the image. This task becomes particularly difficult in the case of textured images. Texture is a term that refers to properties that represent the surface of an object. We might define texture as something consisting of mutually related elements therefore we consider a group of pixels [1]. Image segmentation is the process of partitioning an image into homogenous regions. The existing segmentation methods are commonly classified according to the texture description. The algorithm for texture segmentation deals with first extracting textural features of image by Gabor transform or Discrete Wavelet Transform. The statistical Features can be derived from these images. In case of Wavelet transform Wavelet Pyramid or Wavelet packet Filters are used. In case of proposed method the image is first smoothened and passed through Wavelet pyramid filters, the image is than passed through Histogram filter. Edge detection algorithms are then applied on these images so as to get the boundary of different textures. For non-textured image boundary can be detected by smoothing and then directly applying edge detection algorithm [2].

2. Basic Concepts
2.1 Statistical analysis
Co-occurrence matrix: This method of texture description is based on the repeated occurrence of some Grey level configuration in the texture; this configuration varies rapidly with distance slowly in coarse textures. An occurrence of some grey level configuration may be described by a matrix of relative frequencies P,d (a, b) describing how frequently two pixels with the grey level a, b appear in the window separated by a distance d in the direction . [1]
2.2 Textural Feature Extraction
The filtering approaches to texture classification generally compute the filter output statistics as features. Figure 1 describes a typical scheme of textural feature extraction. The textured image is filtered through a bank of filters tuned to
different frequencies. The filtered image undergoes a nonlinear transform followed by a smoothing operation for output statistics computation as features.
2.2.1 Gabor Filter
Using a Gabor filter bank, an image can be decomposed into orientational components lying in a specified frequency range. This decomposition simplifies higher level image processing like extraction of contours or pattern recognition.Here, the input textural image is filtered through a filter bank having Gabor Filters and Gaussian smoothing filters in cascade. The feature vector is constructed from the output statistics of the images obtained from each branch of the filter bank. Figure 2 [3] illustrates the block diagram for the procedure of textural feature vector construction [3].

Gabor Wavelet Transform
2.2.2 Discrete Wavelet Transform
Wavelet transform is capable of providing the time and frequency information simultaneously, hence giving a time-frequency representation of the signal. A multi-resolution approach is suggested to give a robust segmentation process. This is when an image is decomposed and represented at different scales. The discrete wavelet transform is implemented with a 2-channel analysis filter bank [5]shown in Figure 3. First the low pass filter (intensity) and high pass filter (texture) are applied to the rows of the image. After this stage the columns are down sampled by a factor of 2 (the odd numbered columns are discarded). After this stage the same technique is applied to the 2 resultant images, however this time the filters are applied to the columns of the image and the rows are then down sampled. The resultant is 4 frequency bands, each one quarter of the original size, which makes up the original image.

Wavelet Pyramid Decomposition:
The pyramid wavelet transform recursively decomposes sub-signals in the low frequency channel. This has the effect of concentrating the energy of the image towards the low end of the frequency spectrum, emitting the high frequency information and approaching an approximation of the image [5].

Wavelet Packet (Tree) Decomposition.
Pyramid wavelet transform may not be suitable for quasi-periodic signals, whose dominant frequency channels are located in the middle frequency region [5].



3. Outline Of Propose Method

The algorithm for textured and non-textured images is given. for non-textured images only smoothing and edge detecting algorithm is required, while for detecting edge of textured image the image need to be preprocessed by passing it through wavelet pyramid filter as shown in Figure 4.

Figure: 4 Textured image Edge Detection Scheme

Algorithm For Textured Image Edge Detection
1) Preprocess the image by passing through Low Pass filter.
2) Pass the image through Wavelet Pyramid Filter.
3) Segment the image by passing through Histogram filter.
4) Detect the image by Edge Detection algorithm.

Edge Detection algorithm
1) From difference of Pixel
2) Read the input Pixel.
3) Read another input pixel.
4) Take the difference of the RGB values of two pixels.
5) Display the difference of Pixels.

From Image Difference.
1) Read the image.
2) Smooth the image by applying smoothing Filter.
3) Take the difference of these images.

Histogram filter Algorithm
1) Read image pixels.
2) Convert RGB into HSV Hue saturation value.
3) Find intensity part of HSB.
4) Create Histogram for 255 binary level.
5) if histogram value lies between 10 and 50 make it constant less divide it by 4.
6) display output image

Wavelet Pyramid [5]
1) Read Image.
2) Filter the image through Low pass and high pass Filter.
3) Downsample output of (b) by 2
4) Repeat step (a) & (b) such that low passed image is passed as input of (b).
5) Above step is repeated till a certain level of pyramid.
4. RESULTS
The method has been tested on a large number of various images including synthetic and natural textures. The method of finding edge requires less time as it is simply the difference, rather than applying filter mask like as used in laplacian and sobel operator. Although, this method showed good results it appeared that is was not robust (this is due to the random nature of genetic algorithm). an example of Edge Detection of famous Lena image, the output has taken less calculations output from Typical textured image used image processing,, the result of detecting edges of this from algorithm mentioned for textured image edge detection is also found good.



5. CONCLUSION
This paper has presented a novel algorithm for Edge detection of textured images. The main contribution is on applying a combination of Wavelet pyramid, histogram filter and Edge Detection Filter. The method presented has shown good behavior for both textured and non-textured images. The algorithm for Edge Detection requires less computation than edge detection through masking since the computation required for masking requires convolution. Algorithm for textured image may fail if the textures of images have very low intensity difference. A combination of Gabor Filter along with Histogram filter also can be used to segment multi-textured image.

6. REFERENCES
[1] Anil K. Jain, Fundamentals of Digital Image Processing, Prentice Hall, 1989
[2] Alberto Martin and Sabri Tosunoglu, "Image Processing Technique For Machine Vision", IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 5, February 2000.
[3] Dennis Dunn William E. Higgins and Joseph Wakely, Texture Segmentation using 2-D Gabor Elementary Functions, IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 16, February 1994.
[4] Milan Sonka, Vaclav Hlavac and Roger Boyle, Image Processing, Analysis And Machine Vision, Chapman & Hall 1995.
[5] S.C. Liew, H. Lim, L.K. Kwoh, and G.K. Tay, Texture Analysis Of SAR Images, IGARSS '95, International Volume 2 1995.

---