Abstract- Detecting regions of change in multiple images of the same scene taken at different times is of widespread interest due to a large number of applications in diverse disciplines, including remote sensing, surveillance, medical diagnosis and treatment, civil infrastructure, and underwater sensing. In this paper we propose a method of change detection on different types of images. This method combines the concepts of Change Vector Analysis (CVA), K-Means Algorithm, Otsu Thresholding and Mathematical Morphology.CVA is the initial step which helps to detect the intensity level change in multiple images of the same scene taken at different times. The K-means algorithm is an iterative technique that partitions an image into K clusters. Otsu Algorithm is an unsupervised method of automatic threshold selection for image segmentation. Mathematical morphology is very attractive for image segmentation because it efficiently deals with geometrical descriptions such as size, area, shape, or connectivity that can be considered as segmentation-oriented features.

Index Terms- Change Detection, K-Means Algorithm, Otsu Thresholding, Mathematical Morphology.

INTRODUCTION- Change detection has gained significance over years due to its extensive applicability. The available algorithms for this purpose have been successfully used for region recognition in remotely sensed images.

PREVIOUS WORKS-

Various Change Detection Algorithms have been cited in the literature for improving Change Detection methods.

Change Detection Using Standard Deviation Calculation

This approach is based on the calculation of Standard Deviation for each pixel in the concerned images. On the basis of standard deviation calculation of pixels, the discontinuity among the pixel is recorded. Then the image is segmented into a binary image with a fixed threshold where black pixels signify homogeneous region and white pixels denote discontinuity. By comparing the discontinuous regions the changes are detected in the prescribed images. However, if the scenes under consideration have sufficient amount of varied illumination then this algorithm fails to detect the changed regions properly [1].

Change Detection Using Thresholding

Thresholding is a fundamental technique applied in many image processing applications including Change Detection. It develops an approach for efficient and quantitative evaluation of thresholding algorithms for change detection in a surveillance environment. The technique basically attempts for finding a threshold value, which enables the classification of pixels into different categories. A major weakness of this segmentation mode is that: it generates only two classes. Therefore, this method fails to deal with multichannel images [2].

Change Detection Using Adaptive Threshold Technique

Change detection based segmentation algorithms, threshold the frame difference to form change detection mask. Then the change detection masks are further processed to generate final object masks. It deals with achieving efficient image segmentation by adaptive threshold technique. This algorithm can give satisfying Change Detection results with low computation. The processing speed is high, but it is often not robust. The Change Detection results are suffered from the uncovered background situations; still object situations, light changing, shadow, and noise [3].

Change Detection Using Morphological Filters

It is an unsupervised technique for change detection (CD) in very high geometrical resolution images, which is based on the use of morphological filters. This technique integrates the nonlinear and adaptive properties of the morphological filters with Change Vector Analysis (CVA) procedure. Alternating sequential filters are the most effective and permit the preservation of the geometrical information of the structures in the scene while filtering the homogeneous areas. The weakness of this approach is that the efficiency depends on the Filter size. Also the result is found more satisfactory when dealing with Very High Resolution Images [4].STEPS OF PROPOSED METHOD

In this paper, we have discussed a technique of change detection in different types of images. It is assumed that an adequate pre-processing phase has been applied to the multi-temporal images in

order to make them as more comparable as possible. Proposed method consists of four distinct steps and they are discussed below.

Step 1: Change Vector Analysis

Let us consider two gray scale images f1(x,y) and f2(x,y) acquired over the same area at different times t1 and t2 respectively. Here, each image fk with k = 1, 2, has a size of I × J pixels.

We analyse the Difference Image

D(x,y) =| f1(x,y) - f2(x,y)|

Step 2: Image Segmentation using K-Means Algorithm

Image segmentation is the process of assigning a label to every pixel in an image such that pixels with the same label share certain visual characteristics. Each of the pixels in a region are similar with respect to some characteristic or computed property.

In this step we segment the Difference Image generated in the previous step using K-Means Clustering Algorithm. This algorithm aims at minimizing an objective function, in this case a squared error function. The objective function

,

Here ||xi(j) – cj||2 is a chosen distance measure between a data point xi

(j) and the cluster centre cj , is an indicator of the distance of the n data points from their respective cluster centres [5].

Step 3: Calculation of Threshold Limit

Thresholding is very important task in Change Detection Algorithms. The accuracy of an algorithm is dependent on the choice of threshold parameters.

The criteria of selection of a parameter for a given image are that the resultant should satisfy the following: (i) It should contain most of the prominent changed regions (ii) It should not contain too much unwanted noise (iii) It should be meaningful and visibly pleasing.

In the proposed algorithm we calculated the Threshold limit using Otsu Thresholding.

In Otsu Thresholding we iterate through all the possible threshold values and calculating a measure of spread for the pixel levels Xi each side of the threshold, i.e. the pixels that either fall in foreground or background using histogram Hi of the image D(x,y).

b = ∑ (Xi - µb)2 * Hi

where, 0 i b

f = ∑ (Xi - µf)2 * Hi

where, f i 255

Within Class Variance is calculated as

W = wbb + wf f

The value having lowest sum of weighted variances is the Threshold limit T [6].

Step 4: Thresholding the Segmented Image

In the segmented image generated by applying K-Means Algorithm, those pixels having Intensity value less than Threshold limit found by applying Otsu Thresholding are Zeroed. The pixels having Intensity value greater than Threshold limit are kept as it is.

K(x,y) = 0 if S(x,y) T

= S(x,y) else

Step 5: Morphological Operation

This step is applicable only for Medical Images. Here Morphological Erosion Operator is applied on the image K(x,y) [7][8].

The basic effect of the operator is to erode away the boundaries of regions of foreground pixels. Thus areas of foreground pixels shrink in size, and holes within those areas become larger.

If K(x,y) denotes an input image and Mn denotes a window or flat structuring element of size n , the erosion by the flat structuring element Mn is given by

G(x,y) = K(x,y)) in{ K(x+xo,y+yo),(xo,yo) Є MN }

Step 6: Counting the number of neighbourhood Pixels

In the image generated above, for pixels having intensity value greater than Zero, the number of neighbourhood pixels having intensity value greater than Zero is counted. If the count value is greater than 4, then we set the pixel intensity to 255, else 0.

O(x,y) = 0 if count 4

= 255 else

Figure 1: Methodology

EXPERIMENTAL RESULTS:

The input images are considered to be .pgm images. All the image files that we have tested are natural images. The number of segments obtained are tabulated in Table I and compared to known Change Detection technique.

It has been observed that when Change Vector Analysis (CVA) is combined with K-means algorithm, Otsu Thresholding and Mathematical Morphology, it produces reliable results.

(a) (b) (c) (d)

(e) (f) (g) (h)

Figure 2 (a) Cell Image without Patch (b) Cell Image with patch (c) Image after CVA (d) Segmented Image (e) Image after thresholding Segmented Image (f) Image after Morphological Erosion Operation (g) Final Change Image (h) Ground Truth

(a) (b) (c) (d)

COMPRATIVE STUDY OF DIFFERENT CHANGE DETECTION TECHNIQUES

For Image 1:

(a) (b) (c)

(e) (f) (g)

Figure 3 (a) Hall Image (b) Hall Image with Man (c) Image after CVA (d) Segmented Image (e) Image after thresholding Segmented Image (f) Final Change Image (g) Ground Truth

(d) (e) (f)Figure 3 (a) Cell Image without Patch (b) Cell Image with Patch (c) Reference Image (d) CD using Standard Deviation Calculation (e) CD using Segmentation and Adaptive Thresholding (f) CD using Proposed Algorithm Method Total

Pixels Correct Detection

False Alarm

Missed Alarm

Total Error Pixels Percentage

CD using SD Calculation 16900 pixels

16167 pixels

497 pixels

236 pixels

733 pixels

4.337 %

CD using Segmentation and Adaptive Thresholding

16900 pixels 15925 pixels 108 pixels 867 pixels 975 pixels 5.769 %

CD using Proposed Algorithm 16900 pixels 16118 pixels 189 pixels 593 pixels 782 pixels 4.627 %

Table 02 Comparison of Change detection Techniques for Medical Image 1

For Image 2:

(a) (b) (c)

Method Total Pixels

Correct Detection

False Alarm

Missed Alarm

Total Error Pixels Percentage

CD using SDCalculation 25600 pixels

25290 pixels

195 pixels

115 pixels

310 pixels

1.210 %

CD using Segmentation and Adaptive Thresholding

25600 pixels 25172 pixels 186 pixels 242 pixels

428 pixels

1.671 %

CD using Proposed Algorithm

25600 pixels 25121 pixels

13 pixels 466 pixels 479 pixels 1.871 %

Table 03

Comparison of Change detection Techniques for Medical Image 2

(d) (e) (f)

FIG.22 (a) Cell Image without Patch (b) Cell Image with Patch (c) Reference Image (d) CD using SD (e) CD using Segmentation and Adaptive Thresholding (f) CD using Proposed Algorithm

For Image 3:

(a) (b) (c)

Method Total

Pixels Correct Detection

False Alarm

Missed Alarm

Total Error Pixels Percentage

CD using SD Calculation 262144 pixels

245734 pixels

14711 pixels

1699 pixels

16410 pixels

6.259 %

CD using Segmentation and Adaptive Thresholding

262144 pixels 256258 pixels 3701 pixels 2185 pixels 5886 pixels 2.245 %

CD using Proposed Algorithm

262144 pixels 258431 pixels 27 pixels 3686 pixels 3713 pixels 1.416 %

Table 04

Comparison of Change detection Techniques for Image 3

(d) (e) (f)

FIG.23 (a) Hall Image without Man (b) Hall Image with Man (c) Reference Image (d) CD using SD (e) CD using Segmentation and Adaptive Thresholding (f) CD using Proposed Algorithm

CONCLUSION:

We have proposed a new approach for unsupervised Change Detection technique for gray level image that can successfully detect the change in the images. Comparison of the experimental results with that of other Change Detection methods, show that the technique gives satisfactory results when applied on well known natural and medical images.

REFERENCES:

[1] Soumya Dutta, Dr. Madhurima Chattopadhyay, “A Study of Change Detection algorithm for Medical Cell Images”, International Journal of Information Technology and Knowledge Management, July-December 2011, Volume 4, No. 2, pp. 399-402.

[2] Paul L. Rosin, Efstathios Ioannidis, “Evaluation of global image thresholding for Change Detection”, Pattern Recognition Letters 24 (2003) 2345–2356.

[3] Thrasyvoulos N. Pappas, “An Adaptive Clustering Algorithm for Image Segmentation”, IEEE Transactions on Signal Processing Vol. 10 No. 4 April 1992.

[4] Mauro Dalla Mura, Jon Atli Benediktsson, Francesca Bovolo, and Lorenzo Bruzzone,“An Unsupervised Technique Based on Morphological Filters for Change Detection in Very High Resolution Images”, IEEE Geoscience and Remote Sensing Letters, Vol. 5, No. 3, July 2008.

[5] V.K.Dehariya, S.K.Srivastava,R.C.Jain, “Clustering of Image Data Set Using K-Means and Fuzzy K-Means Algorithms”, Computational Intelligence and Communication Networks(CICN), Nov. 26-28,2010.

[6] Nobuyuki Otsu, “A Threshold Selection Method from Gray-Level Histograms”, IEEE Transactions on Systems, Man, and Cybernetics, Vol. Smc-9, No. 1, January 1979.

[7] Jean Serra, “Image Analysis and Mathematical Morphology”, ISBN 0-12-637240-3 (1982)

[8] L.Vincent, “Morphological grayscale reconstruction in image analysis: applications and efficient algorithms”, IEEE Transactions on Image Processsing , Apr 1993.