Gevşeme temelli kenar belirleme algoritması
Başlık çevirisi mevcut değil.
- Tez No: 75205
- Danışmanlar: DOÇ. DR. TAMER ÖLMEZ
- Tez Türü: Yüksek Lisans
- Konular: Elektrik ve Elektronik Mühendisliği, Electrical and Electronics Engineering
- Anahtar Kelimeler: Belirtilmemiş.
- Yıl: 1998
- Dil: Türkçe
- Üniversite: İstanbul Teknik Üniversitesi
- Enstitü: Fen Bilimleri Enstitüsü
- Ana Bilim Dalı: Biyomedikal Mühendisliği Bilim Dalı
- Bilim Dalı: Belirtilmemiş.
- Sayfa Sayısı: 146
Özet
ÖZET Bu çalışmada esas olarak, görüntü işleme konusunda temel öneme sahip kenar belirleme işlemlerine yer verilmiştir. Ayrıca görüntü iyileştirme konusuna da değinilmektedir. Ele alınan görüntülerdeki bozucu etkiler ve bu etkilerin giderilmesi için kullanılabilecek çeşitli teknikler, ilk olarak anlatılmıştır. Daha sonra çeşitli kenar belirleme algoritmalarına değinilmiştir. Bu tezin ana konusunu oluşturan gevşeme düşüncesi genel olarak anlatılmıştır. Sonra gevşeme düşüncesi ile kurulu metodlardan bahsedilmektedir. En son olarak da kenar belirleme amacıyla oluşturulan gevşeme düşüncesine dayalı bir metod verilmiştir. Metoddan elde edilen deneysel sonuçlar, diğer bazı kenar belirleme algoritmalarının sonuçlan ile karşılaştırılmıştır. IX
Özet (Çeviri)
SUMMARY RELAXATION-BASED AN EDGE DETECTION ALGORITHM Today, digital image processing has a wide area of application. One of the essential subjects is edge detection. Images are composed of many areas with various color levels. Analysis of the contrast difference between these areas is made possible using the edge detection algorithms. Edge detection methods, converting gray level images to edge images, transfer many useful physical properties of the original image without change. An image consisting solely of edges is enough for the original image to be recognized. Thus, enabling the application of data reduced to simple forms especially in the early steps of image processing, edge detection procedures are very important. Edges characterize the borders of the objects in an image. So, it is possible to develop efficient image encoding systems using only the edges in the images. One of the most important problems in image processing is the difference between the original image and the processed image. This difference is mainly caused by blurring and noise. As in all image processing, these factors also affect the edge detection process. Various methods can be used the remove the distortion of blurring and noise. Some these are; average calculation to remove aggregate zero average noise, median filtering to reduce impuls noise, using lowpass filter, image restoration filter or adaptive filter to remove blurring and noise effects. Methods used to remove distortion of the image should satisfy the following conditions: a) Deblurring b) Suppression of noise c) Preservation of discontinuities d) Reducing the ringing effect The edge is the border in an image where some characteristic changes occur following some physical situations like surface reflection, illumination or the distance of the observer from the surfaces in his line of sight. Among these, intensity and color changes are the parametersthat determine an edge. Thus an edge can also be defined as the border or transition area between two levels of different intensity. Edges that can be defined according to their gray level may have different structures. It is useful to know the different edge structures to define the edge detection rules or aggreeable characterization. Typically edges can be characterized as a step function or a ramp between two straight areas. (Figure SI) Actually, these are idealizations of edges. Other edge structures that are seen in real life are shown in Figure S2. a) Step edge b) Ramp edge Figure SI: Some common naive idealizations of edges. Although there may be edges of various structures, an effective edge detection algorithm should be independent from local shape characterization, to reduce the effects of structurel difficulties. An edge detection algorithm should also generally satisfy the following conditions: a)Good detection: True edge point must be marked by the detector and probability of falsely marking non-edge points must be reduced. b)Good localization: Location of detected edges should be as close as possible to true edges. c)Robustness:The algorithm should be robust to noise and perform well for various images. d)Efficiency: The implementation of the algorithm should lead itself to be an efficient one. Parallel form of the algorithm with small neighbor interactions improves the efficiency. e)Applicability to sparse data: The algorithm should reach a reasonable performance when applied to sparse data which allows usage of depth data. The most important difficulty in edge detection algorithms is the dilemma between the location and detection criteria. Correct location of an edge requires processing on the immediate neighbors of the pixel. On the other hand, edge detection criteria requires noise removel in the edge map. But noise reduction and image convolution using a wide standard deviation Gaussian filter needs a wide support area. Another problem is presented by the multidimensional naturel images where objects of different properties and size are located at different distances.Mto the camera. The two dimensional projections of so varying three dimensional objects are likely to change the signal-noise ratio. Thus, the image is distorted. a) Ramp edge. b)Edge between areas of non uniform intensity. rA- c) Step edge between areas of noise. d) Any combination of edge types. e) Line edge. f) Roof edge. g) If these deviations from ideal are severe enough, edge data may be nearly indistinguishable from noise. Figure S2: Cross section of some non-ideal (continuous) edges. During recent work, various methods have been devised for edge detection problems. These methods can be classified as follows: a)Differentiation methods. b)Local statistical methods. c)Stochastic gradient methods. d)Filtering methods. e)Morphological methods. f)Neural network based methods. XIIDifferential methods are most widely used. These in turn are grouped as follows. a)Gradient based methods. b)Compass operators. c)Laplacian based methods. d)Marr-Hildreth's method. Gradient based methods detect edges by calculating the gradient of the image function and evaluating the local extremities. A number of operators are developed for gradient based detection. Most significant are Sobel, Robert, Prewitt and isotropic operators. These operators are applied to the image, then thresholding is used to generate the edge map. Although gradient based methods can be used in detecting all types of edges, the best performance is achieved with step edges. Edges detected with this method resemble a strip rather than a line, thus a final edge thinning may be needed. Performance is affected with high noise levels. Edge detection with compass operators is a slightly changed version of gradient based method. Image is convoluted with eight masks each of which are generated by rotating the previous mask by 45°, thresholding is applied using the biggest gradient and the edge map is generated. Many compass operators have found their way into literature. Compass operator apply edge detection procedures from different angles, and procedure clearer images than gradient based method. Although this method can be used in detecting all types of edges, the best performance is achieved with step edges. Performance is affected with high noise levels. Laplacian based methods are based on calculating the Laplacian of the image function and finding zero crossing. Generally these methods are very sensitive to noise. Local variance conditions may be introduces to prevent over-sensitivity. These methods produce continuous edge lines and edge thinning is not necessary. Absolute value used in the process causes double edge lines. Actually this is not an effective edge detection operator because determining the edge direction presents a problem. But with local variance conditions, it may turn out to be a strong zero crossing detector. Best performance is achieved in images where edge transition is slow. Marr-Hildreth have shown that Gaussian filter, which have spatial and frequency domain location properties, are most appropriate for edge detection. Thus, they developed Laplacian-Gaussian filter. h(x,y)=e-(x2+/)/(2^2) (S.l) XlllH(Qx,fiy)=27iVe-;Kr (fl, +n/)/2 (S. 2) Laplacian-Gaussian filter can be used with any of the partial derivative operators. Laplace operator is generally preferred for its ease of use. After the image is convoluted with the filter, zero crossing are searched. This method does not create gray changes which do not exist in the original image. True edge location is usually quite succesful. Laplacian based methods' sensitivity to noise is avoided with the filtering of noise. Effectiveness of the filter is bound to the size of the central effect area. The smaller the central effect area, the more sensitive is the filter, meanwhile noise tolerance drops. Edges smaller than the central effect area- are detected with high sensitivity, bigger ones are shifted. To detect ramp edges, their width should be smaller than that of the operator. Roof edges are shifted if edge width is smaller than the central effect area. The most important disadvantage of this method is the amount of mathematical operators involved. Generally regarding derivative methods, it is seen that they are incomplete in content. Anyhow, these are the most widely used and most effective methods. Detected edges are usually shifted. Their noise sensitivity is high. Blurred images present a problem. They usually causes discontinuities in the edge maps. Local statistical methods use local average and variation changes to detect edges. Although in low noise environments these methods produce worse results than derivative methods, they can produce a draft edge image even when noise level is very high. In stochastic gradient methods, noise is taken into consideration. In these methods, linear estimate methods are applied to the stochastic models of the image function. Stochastic gradient methods produce results better than the other methods when noise level is high. That is because, mask parameters are calculated according to noise and blur properties and the masks are multidimensional. If multidimensional masks are used in the other methods their performance will also enhance. Another method that might be used with high noise level is filtering methods. In these methods, filters that maximize the energy of edge areas in a defined resolution range are designed, and edge image is produced. Although results with high noise levels are satisfactory, those with low noise levels are not so well. There are few works on the relatively new subject, morphological methods. Edge structures are important here and methods are built upon edge structures. Although results with low noise levels are satisfactory, those with high noise levels are not so. Edges appear to XIVbe shifted from their true locations. These methods have difficulty also with ramp edges. Edge detecting algorithms developed using neural networks have been more significant with the growing importance of neural networks. The edge detection algorithms mentioned above usually do not take observatory data and local neighborhood relations into consideration. They are hard to arrange for general labeling problems. Edge types and structural data about vertices where edges connect are ignored, all possible edge and vertex structures are assumed to be natural and consistent, which is not true. Weak edges usually appear after the widely used differential operations. The objects in the image may not be properly defined. This situation usually arises with natural view images. Relaxatian methods are different from other edge detection methods; they take rich observatory data and local neighborhood relations into consideration. They also use some predefined knowledge to determine he consistent ones among all possible edges and vertices. Many researchers are working on this subject, yet complete works are few. Methods based on the relaxation concept generally reach consistent structures reducing uncertainties, though some of the results may be insufficient, inefficient and uncertain. They reduce noise effects and are succesfully applied to distorted step edges. Good performance may be achieved in high noise levels without using filters that need a wide support area. Researchers whose used relaxation methods neglected directional data despite the fact that they had used observatory data and neighborhood relations. Therefore inconsistant edge structures may appear. In this thesis, an edge detection algorithm based on a relaxation method which includes directional data has been developed and its performance has been tried out. The algorithm steps are as follows: a)Restoration of image. b)Relaxation. c)Edge thinning. Primarily, a preprocess which is to restorate images to experiment images. These results have been tried to be obtained: a)Within a homogeneous region, it smooths out small amounts of noise. b)Near a region boundary whose contrast is greater than threshold it includes no points accross the boundary in the average. This allows a smoothing of the points on either side on the boundary without blurring the boundary as a nondiscriminatory averaging process would do. XVc)Within an intensity gradient, the process averages a point with roughly as many other points that have smaller intensity as greater. This will smooth noise within the gradient, but it will not destroy the gradient. d)In a textured region, if the texture elements differ little in intensity, they will be smoothed into a homogenous region. If the texture elements differ by more than threshold, then no averaging will occur, except perhaps within the texture elements themselves. In the relaxation step of the algorithm, there is a method that uses the directional data. Thus, the disadvantages of the methods which ignore the directional data have been tried to be removed. The last step of the algorithm is edge thinning.. The skeleton of the edge-map has been tried to be made this process. It has been observed that successful results have been obtained when the developed algorithm is applied to the camera images with the application of the directional data. The algorithm is also robust to noise. In general, this algorithm is faster compared to the other relaxation-based methods. XVI
Benzer Tezler
- Sanal gerçeklik temelli gevşeme programının endoskopi uygulanacak hastalarda ağrı şiddetine, anksiyete düzeyine ve hasta memnuniyetine etkisi
The effect of virtual reality-based relaxationprogram on pain intensity, anxiety level andpatient satisfaction in patients to BE appliedwith endoscopy
KÜBRA ASLAN
Yüksek Lisans
Türkçe
2023
Hemşirelikİstanbul Sabahattin Zaim ÜniversitesiHemşirelik Ana Bilim Dalı
DR. ÖĞR. ÜYESİ ZÜLFÜNAZ ÖZER
DOÇ. DR. MUSTAFA KEMAL YÖNTEM
- Sanal gerçeklik temelli gevşeme programının bronkoskopi uygulanacak hastalarda ağrı şiddetine, anksiyete düzeyine ve hasta memnuniyetine etkisi
Başlık çevirisi yok
AYŞEGÜL SARIKÖSE
Yüksek Lisans
Türkçe
2023
HemşirelikFırat Üniversitesiİç Hastalıkları Ana Bilim Dalı
DR. ÖĞR. ÜYESİ GÜLCAN BAHÇECİOĞLU TURAN
- Eğitim ve progresif gevşeme egzersizleri temelli bir uygulamanın zihinsel engelli bireylerin bakım vericilerinin yaşam kalitesi, algıladıkları stres ve başa çıkma yolları üzerine etkisi: Randomize kontrollü çalışma
The effect of education and progressive relaxation exercise-based practice on quality of life, perceived stress and coping ways of caregivers of individuals with intellectual disabilities: A randomized controlled study
GÖNÜL GÖKÇAY
- Akut miyokard infarktüsü sonrası uygulanan bilinçli farkındalık temelli gevşeme uygulamasının yaşam kalitesi, yorgunluk ve hareket korkusuna etkisi: randomize kontrollü çalışma
The effect of mindfulness meditation on quality of life, fatigue and kinesiophobia among patients with acute myocardial infarction: a randomized controlled study
CANAN KARADAŞ
Doktora
Türkçe
2020
HemşirelikHacettepe Üniversitesiİç Hastalıkları Hemşireliği Ana Bilim Dalı
PROF. DR. LEYLA ÖZDEMİR
- Covid-19 pandemi döneminde epilepsi tanılı çocuklara evde uygulanan gevşeme ve solunum egzersizlerinin yaşam kalitesi, anksiyete, depresyon ve uyku kalitesine etkisi
The effect of relaxation and breathing exercises applied at home to children with epilepsy on the quality of life, anxiety, depression and sleep quality during the covid 19 pandemic period
SÜHEDA HASÖZDEMİR
Yüksek Lisans
Türkçe
2023
Çocuk Sağlığı ve Hastalıklarıİstanbul ÜniversitesiPediatrik Temel Bilimler Ana Bilim Dalı
PROF. DR. MELİHA MİNE ÇALIŞKAN