Dijital Görüntü İşleme Teknikleri Kullanılarak Görüntülerden Detay Çıkarımı

Abstract

Günümüzde bilgisayar teknolojisinin gelişmesi ile birlikte dijital görüntü işleme teknikleri yaygın bir şekilde kullanılmaktadır. Dijital görüntü işleme uzaktan algılamanın uygulamaları arasına girmektedir. Dijital görüntü işleme çevreyle hiçbir etkileşim olmaksızın teknolojik araçlar yardımıyla görüntüler içerisinde bulunan objelerin öznitelikleri hakkında bilgi edinmemizi sağlar. Dijital görüntü işleme teknikleri ile iyileştirilmiş veya daha farklı görüntüler elde edilmekle birlikte görüntülerden öznitelik ve anlamlı bilgi çıkarımı yapılabilmektedir. Dijital görüntüler sayesinde görüntü yorumlamaya olanaklı olan mekansal veriler, kentsel planlama, çevresel değişimin incelenmesi, coğrafi analiz verilerinden bilgisayar teknolojisi yardımıyla birçok analiz yapılmaktadır. Bu işlemler sayesinde şu ana kadar birçok yaklaşım ve yöntem sunulmuştur. Bulunan yöntemlerin görüntülere özgü yöntemler olmasından dolayı görüntü işleme konusu birçok farklı türdeki görüntüler için ayrı ayrı yöntemler olarak sunulmuştur. Bu çalışma kapsamında birinci bölümde dijital görüntü işlemenin tarihçesi ve uygulama alanları hakkında bilgi verilmiştir. İkinci bölümde, görüntü elde etme ve görüntü işleme tekniklerinin temel kavramlar üzerinden örnek görüntüler verilerek anlatılmıştır. Özellikle bu tez çalışmasında, en yaygın kullanılan görüntü işleme tekniklerine değinilmiştir. Üçüncü bölümde MATLAB programlama dilinin avantajlarına değinilmiştir. Kod yazma ortamı ve arayüz tasarlama ortamı olan ‘Guide’ araç fonksiyonu tanıtılmıştır. Uygulama bölümünde, tez çalışmasında kullanılan görüntülerden ve uygulanan yöntemlerden bahsedilmiştir. Nokta işleme teknikleriyle görüntüler hakkında bilgi sağlamakla birlikte kontrast ve parlaklık ayarlarına değinilmiştir. Çeşitli filtreler geçirilerek görüntü zenginleştirme yöntemleri denenmiştir. Görüntülerin gri seviyelerindeki ani değişikliklerin olduğu bölgelerde kenar belirleme algoritmaları kullanılmıştır. Bu kenar algoritmalarından birinci türeve ve ikinci türeve dayalı kenar belirleme algoritmalarından belli eşik değer altında çıkarılan detaylar yorumlanmıştır. Morfolojik yöntemlerin çeşitli operatörleri kullanılmıştır ve daha önceki çalışmalara bu operatörler eklenerek detay ve sınırların çıkarılması çalışılmıştır. Bu kapsamda dijital görüntü işleme tekniklerinin klasik yapısal programlama dilleri kullanılarak gerçekleştirilmesi; yoğun matematiksel işlemlerin olması, veri sayısının yüksek ve algoritma karışıklığının olması gibi nedenlerden dolayı zorlaşmaktadır. Bu sorunu aşmak için MATLAB programlama dili tercih edilmiştir. MATLAB’in dijital görüntü işleme fonksiyonları kullanılarak görüntü işlemeye yönelik uygulama arayüz tasarımı gerçekleştirilmiştir. Son olarak beşinci bölümde değerlendirmeler ve öneriler verilmiştir.

Today, with the development of computer technology, digital image processing techniques are being widely used. Digital image processing is one of the remote sensing applications. The availability of digital image processing techniques have significantly enhanced the possibilities for photogrammetric image measurement and analysis. As in other fields, digital images not only enable new methods for the acquisition, storage, archiving and output of images, but most importantly for the automated processing of the images themselves. Digital image processing without any interaction to environment allows us to obtain information about the features of objects that contained inside of the images by means of technological devices. By digital image processing, improved or with more different images that have been obtained, it can be carried out meaningful information and feature extraction from images. Owing to the digital images, many analysis are being carried out on spatial data; which give the possibility of interpretation in images, urban planning, examination of environmental change, geographical data analysis thanks to computer technology. Owing to these processes, many approaches and methods have been put forward. Due to the found methods that related to the specific image methods, image processing subject has been presented for the many different types of images as separated methods. Within the scope of this thesis, information about the history of digital image processing and application areas of have been given in the first section. In the second part, image acquisition and image processing techniques are described in the sample images through the basic concepts. Especially in this master thesis, there has been focused on the most commonly used image processing techniques. Digital image processing techniques, is divided into three parts in this thesis. First part is point operations. These are histogram, histogram equalization, contrast stretching and thresholding. Histogram provides the frequency distribution of the pixel values in the image. It displays the absolute or relative frequency of each pixel value either in tabular or graphical form. The most important parameters of a histogram are; relative and absolute frequencies, minimum and maximum pixel value of the image, contrast and mean of pixel values. While minimum and maximum pixel values define the image contrast, the mean is a measure of the average intensity of the image. Histogram equalization is a non-linear process aimed to highlight image brightness in a way particularly suited to human visual analysis. The cumulative frequency function is calculated from the histogram of the original image. Contrast stretching is linear interpolation between gmin and gmax. Minumum and maximum pixel values can be derived from the histogram, or defined interactively. Thresholding is used to clearly differentiate pixel values which belong to different object classes to separate object and background. Thresholding is often a pre-processing stage prior to segmentation. The second part of techniques is image enhancement methods. These methods are image smoothing, image sharpening and edge detection. In image processing, many filter operations are applied to an image by performing a special operation called convolution with a matrix called a kernel. Kernels are typically 3x3 square matrices, although kernels of size 2x2, 4x4, and 5x5 sometimes used. The values stored in the kernel directly relate to the results of applying the filter, and filters are characterized solely by their kernel matrix. Convolution can be described as a function that is the integral or summation of two component functions, and that measures the amount of overlap as a function is shifted over the other. Filters can be used for denoising signal and images. Many different filters can achieve this purpose and the optimal filter often depends on the particular requirements of the application. One such filter is called Gaussian, so named because the filter’s kernel is a discrete approximation of the Gaussian distribution. The Gaussian filter is known as a ‘smoothing’ operator, as its convolution with an image averages the pixels in the image, affectively decreasing the difference in value between neighboring pixels. The σ parameter is standard deviation of the Gaussian, and be adjusted according to the desired distribution. Another smoothing filter is average filter. Average filter is reducing the amount of intensity variation between one pixel and the next. It is often used to reduce noise in images. The idea of mean filtering is simply to replace each pixel value in an image with the mean (`average') value of its neighbors, including itself. This has the effect of eliminating pixel values which are unrepresentative of their surroundings. Mean filtering is usually thought of as a convolution filter. Like other convolutions it is based around a kernel, which represents the shape and size of the neighborhood to be sampled when calculating the mean. The other filter is median filter. The non-linear median filter performs good smoothing while retaining sharp edges. The median filter is not based on convolution. Instead the median value (as opposed to the mean) output pixel value. The output image therefore consists only of pixel values which exist in the input image. This property is essential for the filtering of images which consist of attributes or special palettes instead of intensities. The median filter is a member of the group of rank-order filters. The median is the centre of a rank-ordered distribution. Laplacian operator is the one of the image sharpening filters. Laplacian operator is a template which implements second order differencing. The second order differential can be approximated by the difference between two adjacent first order differences. The other technique of the image enhancement is edge detection. Many approaches to image interpretation are based on edges, since analysis based on edge detection in insensitive to change in the overall illumination level. Edge detection highlights image contrast. Detecting contrast, which is difference in intensity, can emphasize the boundaries of features within an image, since this is where image contrast occurs. There are two differential filters in the edge detection. Taylor series analysis reveals that differencing adjacent points provides an estimate of the first order derivative at a point. An alternative to taking the maximum is simply to add the results of the two templates together to combine horizontal and vertical edges. There are of course more varieties of edges and it is often better to consider the two templates as providing components of an edge vector: the strength of the edge along the horizontal and vertical axes. These give components of a vector and can be added in a vectorial manner. The edge magnitude is the length of the vector, and the edge direction is the vector’s orientation. The first order differential edge filters are Sobel and Prewitt edge filters. Canny edge detection aims to reduce the response to noise. This can be affected by optimal smoothing; Canny was the first to demonstrate that Gaussian filtering is optimal for edge detection. Moreover second order differential filter is Laplacian of Gauss filter. It illustrates the sensitivity to noise of the Laplacian filter, hence minor intensity changes are interpreted as edges. The second derivative of the Gaussian function is regarded as an optimal edge filter which combines smoothing properties with edge extraction capabilities. The third part of techniques is Morphology operations. Morphology was originally developed for binary images and was extended to include grey-level data. The word morphology concerns shapes: in mathematical morphology we process images according to shape, by treating both as sets of points. There are several morphology operators. These are erosion, dilation, opening, closing. Erosion, which leads to the shrinking of regions. Dilation, which yields to the extension of connected regions. Opening, achieved by an erosion followed by dilation. Small objects are removed. Closing, the reverse of opening. Dilation is followed by erosion in order to close gaps between objects. The third part has been referred to the advantage of the MATLAB programming language. As a code-write and interface design space, 'Guide' tool function has been introduced. The images used and the methods applied in this thesis are mentioned in the fourth section. With the point processing technique, information about the images is provided. In addition to this, contrast and brightness settings have been experienced. Image enhancement methods have been tried through various filters processes. Edge detection algorithms have been used in areas where there are sudden changes in gray level images. Features derived under a certain threshold of edge defining algorithms based on the first and second derivative of all these edge algorithms have been interpreted. Various operators of morphological methods have been used and by adding these operators to the previous studies, extraction of boundaries and features have been studied. Within this scope, due to reasons like the higher number of data and algorithmic complexity, intensive mathematical process, classic structure of language programming of digital image processing techniques is difficult. In order to cope with these challenges, MATLAB programming language has been preferred. Application interface is designed for digital image processing techniques with using MATLAB image processing toolbox. And finally, in the fifth section, evaluations and recommendations have been given.