International Journal Of Communication And Computer Technologies

Volume 01 No.1, Issue: 01 JULY 2012

SIMULATION OF AN AUTOMATIC IMAGE REGISTRATION THROUGH HISTOGRAM BASED IMAGE SEGMENTATION

T.DEVIKA*, Mrs.G.VALARMATHI** *M.E (APPLIED ELECTRONICS)-FINAL YEAR **ASSISTANT PROFESSOR GNANAMANI COLLEGE OF TECHNOLOGY

ABSTRACT: Automatic image registration is still an actual challenge in several fields. Although several methods for automatic image registration have been proposed in the last few years, it is still far from a broad use in several applications, such as in remote sensing. In this paper, a method for automatic image registration through histogram-based image segmentation (HAIRIS) is proposed. This new approach mainly consists in combining several segmentations of the pair of images to be registered, according to a relaxation parameter on the histogram modes delineation, followed by a consistent characterization of the extracted objects through the objects area, ratio between the axis of the adjust ellipse, perimeter and fractal dimension and a robust statistical based procedure for objects matching. The application of the proposed methodology is illustrated to simulated rotation and translation. HAIRIS allows for the registration of pairs of images (multitemporal and multisensor) with differences in rotation and translation, with small differences in the spectral content, leading to a subpixel accuracy. INTRODUCTION
AUTOMATIC IMAGE REGISTRATION

image processing related applications. Remote sensing applications are one of the fields where further research on AIR methods is required. Under this scope, there are particular difficulties so that AIR methods suitable for many computer vision applications will present limited performance. The rigid-body model under the scope of automatic image registration methods is still a present subject of research, in particular under the scope of remote sensing applications. The problem of registering remote sensing images can roughly be locally seen as the determination of translations and a small rotation. Under the scope of computer vision applications, the rigid-body transformation may seem a simple problem to solve with many existing methods. Under the scope of remote sensing applications, one of the major problems is related to the radiometric content (due to multi sensor or multispectral pairs of images). Moreover, scale is frequently known, as most satellite images are provided with sufficiently accurate scale information, being the exception aerial photographs when the aircraft is flying across a region with significant differences in the terrain elevation.

Automatic image registration (AIR) is still a present challenge regarding

Page 13
International Journal Of Communication And Computer Technologies

www.ijccts.org

International Journal Of Communication And Computer Technologies

Volume 01 No.1, Issue: 01 JULY 2012

HISTOGRAM BASED METHODS Histogram-based methods are very efficient when compared to other image segmentation methods because they typically require only one pass through the pixels. In this technique, a histogram is computed from all of the pixels in the image, and the peaks and valleys in the histogram are used to locate the clusters in the image. Color or intensity can be used as the measure. A refinement of this technique is to recursively apply the histogram-seeking method to clusters in the image in order to divide them into smaller clusters. This is repeated with smaller and smaller clusters until no more clusters are formed. Histogram-based approaches can also be quickly adapted to occur over multiple frames, while maintaining their single pass efficiency. The histogram can be done in multiple fashions when multiple frames are considered. The same approach that is taken with one frame can be applied to multiple, and after the results are merged, peaks and valleys that were previously difficult to identify are more likely to be distinguishable. The histogram can also be applied on a per pixel basis where the information results are used to determine the most frequent color for the pixel location. This approach segments based on active objects and a static environment, resulting in a different type of segmentation useful in Video tracking. Histogram-based image segmentation comprises three stages: Recognizing the modes of the histogram, Finding the valleys between the identified modes

Finally apply thresholds to the image based upon the valleys. SCOPE OF THE PROJECT A method for automatic image registration through histogram-based image segmentation (HAIRIS) is proposed, which allows for a more detailed histogram-based segmentation, rather than the traditional methods, and consequently to an accurate image registration. HAIRIS is able to estimate the rotation and/or translation between two images which may be multitemporal or multisensory with small differences in the spectral content. PROJECT DESCRIPTION: HAIRIS DESCRIPTION: Finding the relationship between two coordinate systems using pairs of measurements of the coordinates of a number of points in both systems is a classic photogrammetric task. The transformation between two Cartesian coordinate systems can be thought of as the result of a rigid-body motion and can, thus, be decomposed into a rotation and a translation. Suppose that we have a pair of images in the same scale (same pixel size with respect to the scene), and the existence of a translation and/or rotation difference between the two images, where one of them is static (image 1) and the other (image 2) is to be registered onto the static image. Assuming that are the

coordinates of the static image and are the (Pixel, Line) pair of the image to be
Page 14

International Journal Of Communication And Computer Technologies

www.ijccts.org

International Journal Of Communication And Computer Technologies

Volume 01 No.1, Issue: 01 JULY 2012

registered. The transformation may be written as where the origin is considered to be the upper left corner of the static image, is the orientation difference, and is the shift between the two images. The proposed methodology of automatic image registration is schematically represented in Fig. 1. It begins with a pre-processing stage in order to reduce unnecessary detail on the images content, important for the subsequent histogram-based image segmentation phase (which includes a relaxation parameter). The objects extracted from the segmentation stage are characterized and matched according to some related properties, which finally allows for the statistically-based rotation and translation parameters estimation. In the following, the several steps involved in HAIRIS are explained, where the main objective is to estimate. PREPROCESSING: Too much detail on the pixel domain may lead to undesirable segmentation results. Therefore, it is advisable an image Enhancement step prior further processing. By image enhancement (which is itself a largely subjective process), it is intended to obtain an image with less detail than the original version, nearest to the object identification which is performed by the human eye. Although typically more mathematical and complex, restoration algorithms may provide the exploitation of the detailed characteristics of an image and its degradation. Despite the main purpose of image restoration methods is to model and remove the degradation, these methods may also be used with other purposes. Since with image segmentation it is intended to extract objects (in particular their boundaries), one may view the image

objects which have some texture as a kind of degradation. Therefore, it is intended to remove that degradation, which is assumed to be additive random noise. The Wiener filter is one of the most used filters under the scope of image restoration methods. However, it may also be used for image enhancement, with the aim of reducing the detail on an image, since it is typically a low pass filter and consequently induces a significant blurring effect. Its frequency response is given by where stands for the original image power spectra and for the additive random noise power spectra. Although one might be interested in inducing some blurring on the image, the use of a fixed (space-invariant) filter throughout the image may decrease the clearness of the objects boundaries. Therefore, an adaptive image restoration is required. A pixel-by-pixel processing approach may become quite computationally expensive, in opposition to a sub image-by-sub image processing where we divide the image into a certain number of tiles. The latter is typically considered for sub images with size between 8X8 and 32X32 pixels. However, since the objective of the Wiener filter employment is different from restoration, it is advisable to consider the conservative smallest possible square tile (beyond a single pixel) size of 3X3 pixels. Although this latter approach might induce the socalled blocking effect, it may be ignored for images with low SNR. Recalling the additive random noise, we generally have no a priori information, which may be overcome by considering measured features such as the local variance, providing the determination
Page 15

International Journal Of Communication And Computer Technologies

www.ijccts.org

International Journal Of Communication And Computer Technologies

Volume 01 No.1, Issue: 01 JULY 2012

of the presence of significant highfrequency details. Since we are interested in delineating the objects boundaries, an edge-sensitive adaptive image restoration version of the Wiener filter is adequate. This method is based upon the idea of reducing more noise near edges without additional edge blurring through a cascade of 1-D adaptive filters. Let Ti [.](1<=i<=N) represent a 1-D (space variant) filter, where N is four and represents the four directions corresponding to the angles of 0deg, 45deg, 90deg, and 135deg. Then, these operators are sequentially applied, which lead to an improvement over the performance of some 2-D adaptive restoration algorithms, and to typically less computational requirements. Additionally, in order to overcome significant differences between the histograms of the images to be registered, an histogram equalization of image 2 using the histogram counts of image 1 is performed, prior to the application of the Wiener filter. In this way, the Wiener filtering on image 2 allows both for the reduction of the image detail, as well as to the smoothing of the histogram, which becomes spiky due to the histogram equalization step.
HISTOGRAM BASED SEGMENTATION:

the sequence of the consecutive slopes, where M+1 is the number of histogram levels ( M=255 for an 8-b image). The idea behind this approach is to choose an adequate threshold for considering whether or not one is in the presence of a mode, which is characterized by a significant increase and/or decrease on the slopes sequence. In Fig. , the histogram of the image in Fig is shown, where roughly seven modes are visible. Also in the same figure, the consecutive slopes of the histogram are shown in fig 6.3, where each global transition from positive slopes to negative slopes is associated to a mode on the histogram.

MODE DELINEATION AND IMAGE SEGMENTATION: The method used for mode delineation is based upon the analysis of the consecutive slopes of the histogram. Let x (m) be the image histogram counts m=0 ., M and

(a) Input Image (b) Histogram of the image

y n=1..,M

(n)=x(n)-x(n-1)

Page 16
International Journal Of Communication And Computer Technologies

www.ijccts.org

International Journal Of Communication And Computer Technologies

Volume 01 No.1, Issue: 01 JULY 2012

a class. Then, for each class, those pixels which are 4-connected are considering as belonging to the same object, resulting in the final segmentation of the image. In some cases, a significant amount of no data (DN=0orDN=255 for unsigned 8-b images) may be present, which can mask in some way the histogram shape. Therefore, those corresponding pixels should be discarded prior to the representation of the image histogram. CHARACTERIZATION OF THE EXTRACTED OBJECTS: The extracted objects at the segmentation stage are characterized by four attributes which allow for their adequate morphological description: area (Area), perimeter (Perim), axis ratio (ARat) and fractal dimension (Db) . The attribute area is merely obtained by the number of pixels which form an object, whereas the perimeter is obtained by calculating the distance between each adjoining pair of pixels around the border of the region. These two attributes allow for the evaluation of an object with respect to its size and compactness, respectively. The major axis length corresponds to the major axis length of the ellipse that has the same normalized second central moments as the object, from which it may be also obtained the minor axis length. The ratio between the major and the minor axis length lead to the attribute (axis ratio). The attribute allows for the characterization of the object according to its narrow or wide nature. Although the three previously described measures comprise the most general aspects of an object characterization, several uncertainties would still persist on the objects matching stage. Therefore, there is the need for considering a complementary attribute
Page 17
International Journal Of Communication And Computer Technologies

Slopes of the histogram A relaxation parameter () is considered on the mode delineation, which in theory is a continuous parameter, defined on the space [0, 1], and has to pass through a discretization process in practice. The inclusion of this parameter leads to the obtention of several different segmentation results, which allows for the subsequent stages of the proposed methodology to be more robust. The relaxation parameter () corresponds to the proportion of the height of the histogramconsidered to correspond to the highest modefor which below this value, the mode is to be considered as a flat region. In this way, the method becomes much more useful and adaptable to a large variety of situations. The common methods of merely identifying a mode correspond to selecting the null value for (). Once the modes and flat regions are identified for each value of (), the image may be segmented considering each interval (either a mode or a flat region) as

www.ijccts.org

International Journal Of Communication And Computer Technologies

Volume 01 No.1, Issue: 01 JULY 2012

which considers the particular complexity nature of an object shape: the fractal dimension. Fractal dimension is one among several notions of dimension proposed by mathematicians. In this work, the box-counting dimension Db one of the special forms of Mandelbrots fractal dimension was the considered fractal dimension. It generally consists on the slope of a straight line, fitted to a scatter plot with log(N(s)) and log (1/s) on the vertical and horizontal axis, respectively, where is the mesh size of the grid overlaid on the object and N is the number of grid boxes which contain pixels of the object. The four previously described attributes are used for the later stage of objects matching. These four attributes are expected to be similar for corresponding objects. As stated at the beginning of HAIRIS description, it is assumed that the pair of images differ with respect to rotation ( ) and translation ( x and y). Therefore, in order to allow for the registration of the pair of images, there is the need to have an orientation and positioning indicators. Regarding orientation, for each extracted object, the angle between the x-axis and the major axis of the ellipse that has the same second-moments as the object is stored. With respect to translation, the centroid of each object (the center of mass of the object) is also stored. MATCHING: The matching step begins with the evaluation of a cost function, between every possible two-by-two combination of objects obtained by the segmentation of the two images, for every possible combination of the values considered for both images. This leads to a matrix with rows and columns, where and correspond to the number of extracted objects from images 1 and 2, respectively. The cost

function, evaluated for the values of the properties of the objects from images 1 and 2, is defined as follows:

Where and are the average of each property for images 1 and 2 values. With defined in this manner, it allows for a normalized sum of the considered properties, and consequently to the sum of them. Then, the values are represented in the form of box plots, with the image which led to the lower number of segmented objects corresponding to the horizontal (categorical) axis. A valid matching between two objects should lead to the lower values of, sufficiently far from the majority. This can be statistically evaluated through the outlier detection criterion used in the box plots representation, where a point is considered an outlier (regarding the smaller values) if it is smaller than, where and are the first and third quartiles, respectively. Although is typically considered as 1.5, in this step the more flexible value of 1 is required (also commonly used in practice), in order to reduce the loss of eventual matching candidates. This procedure of outlier detection is applied to each object of the image in the horizontal axis. Segment of the images shown in Fig. where (a) corresponds to band 1 and (b) to band 2 with the simulated shift. Superimposed (in white) are the boundaries of two objects obtained from the segmentation stage, which have been matched. The attributes values from the
Page 18

International Journal Of Communication And Computer Technologies

www.ijccts.org

International Journal Of Communication And Computer Technologies

Volume 01 No.1, Issue: 01 JULY 2012

object in (a.1) are (1759, 2.66, 647.7, 1.12) and from the object in (b.1) are (1685, 2.67, 555.2, 1.09).

allows for the detection of a modal class, restricting the set of possible values for rotation. Then, among these rotation candidates, a robust strategy of finding is performed, through considering the frequencies of the rotation candidates, and finding the rotation value which absolute frequency corresponds to the higher outlier, according to the procedure of box plot outliers detection previously described. This procedure leads to a robust estimation . TRANSLATION ESTIMATION: Once is obtained, only the initial matching candidates which correspond to the obtained rotation are considered. Then, a similar procedure as that followed in the rotation estimation is considered for obtaining (x and y). CONCLUSION A large variety of automatic image registration methods can be found. In several applications, the registration model only assumes rotation and translation, where the registration of satellite images is an example. In this paper, a new approach for automatic image registration through Histogram-based image segmentation (HAIRIS) is proposed, The presented images were contrast stretched in order to allow for a better visualization. Clear advantages by joining these two main areas of image processing. Since HAIRIS does not require any search interval either for rotation or translation, it is a fully automatic procedure.

(a) Image (b) Segment of the image ROTATION ESTIMATION: The rotation and translation are determined on a statistical basis. Given the set of matching objects candidates, the histogram of the extracted objects orientation differences is represented. This

Page 19
International Journal Of Communication And Computer Technologies

www.ijccts.org