Postupci detekcije ruba pri obradi digitalne slike

Cilj ovog završnog rada bio je analizirati i usporediti razne algoritme za detekciju rubova te implementirati iste. U ovom završnom radu definirani su i analizirani Prewitt, Sobel, Robert’s cross, Canny, Laplasijan i Gausov Laplasijan. Prva tri detektora ruba spadaju u metodu pronalaženja gradijenta i koriste prvu derivaciju za detektiranje rubova. Zadnja dva detektora koriste drugu derivaciju za detektiranje rubova i spadaju u Laplasijan detektore. Prewitt i Sobel koriste se za detekciju rubova u vertikalnom i horizonatalnom smjeru, a Robert’s cross koristi se za detekciju rubova pod kutem od 45 stupnjeva. Ukoliko želimo uz usrednjavanje gradijenta zadržati i što više detalja, koristimo Sobel detektor umjesto Prewitta. Ako radimo sa slikom na kojoj ima puno smetnji, Prewitt je malo bolji od Sobel detektora i puno bolji od Robert’s cross detektora. Za Canny možemo reći da je poboljšana verzija nekog od prethodna tri detektora, iako se najčešće kao temelj uzima Sobel. Na temeljni algoritam se dodaju još neke operacije koje omogućavaju precizniju detekciju i smanjivanje debljine rubova. Laplasijan i Gausov Laplasijan bolji su u odnosu na ostale navedene algoritme po pitanju brzine izvršavanja jer imaju samo jednu konvolucijsku matricu (ostali imaju dvije).The purpse of this final thesis was to analyze and compare various edge detection algorithms and to implement them. Prewitt, Sobel, Robert’s cross, Canny, Laplacian and Laplacian of Gaussian edge detectors were defined and analyzed in this thesis. First three are based on gradient method and they are first derivative edge detectors, last two are based on Laplacian method and they are second derivative edge detectors. Prewitt and Sobel are used for edge detection in vertical and horizontal direction and Robert’s cross is used for detecting an edge with angle of 45 degrees. If with averaging gradient magnitude we want to keep the details too, we need to use Sobel detector instead of Prewitt. If we are processing a picture with a lot of noise, Prewitt will be a little bit better then Sobel detector and a lot better than Robert’s cross detector. For Canny we can say that it is improved version of some of these three edge detectors, even Sobel is most used as a basis for Canny detector. On that basis algorithm Canny adds a few more operations which add more precision for detecting an edge and also provides edge thinning. Laplacian and Gaussian of Laplacian have only one convolution matrix (and other edge detecion algorithms have two) and that is the reason they are faster than the others