Vessel segmentation based on Sobel operator and fuzzy reasoning

Abstract

Vessel detection is a fundamental step in retinal analysis which helps to extract further information such as characterization of changes in blood vessels width and tortuosity. In this paper, we present an automatic algorithm based on edge detection and fuzzy inference. In the proposed method, the direction of linear structures are determined with Radon transform, then Sobel operator is used for extracting edges along with the predetermined direction. The combination of gradient information of vessel edges with fuzzy theory and genetic algorithm help us in the vessel validation process. Ultimately, the vessel is reconstructed via the extracted edges and morphological algorithms. Experimental results on 40 images of DRIVE database show that the proposed algorithm despite its simplicity has a high performance in comparison with other edge detector algorithms which are found to be sensitive to noise. Keywords-Sobel operator; Radon transform; Fuzzy inference; Genetic algorithm I. INTRUDUCTIONMany diseases such as Diabetes and hypertensive retinopathy can be diagnosed by investigating retinal images for various features of vessels such as width, length, color, and tortuosity [1]. A good vessel map can provide useful geometric information about the detection of other objects in retina images such as exude, macular and fovea. Thus, a reliable method of vessel detection and quantification would be valuable.Retinal vessels can be described as a dark object set against a lighter background with almost parallel edges. Vessels are piecewise linear structures and intensity profile of them is an approximation of a Gaussian or mixture Gaussian. These factors help us to detect vessels properly but there are some factors which make this process somehow difficult. Shape, size and the color of all vessels are not the same. Regarding these factors different techniques has been suggested for extracting vessels from retinal images.Vessel detection algorithms can be divided into several categories such as locally adaptive scheme methods, pattern recognition methods, tracking based methods, model based approaches and edge based methods. Among these various methods, pattern recognition methods and edge based methods are more important.Techniques based on vessel centerline fall into the category of pattern recognition. In these methods vessel centerline is detected by means of algorithms such as ridgebased algorithms [2][3][4][5], algorithms based on differential geometry [6,7] or thinning based algorithms [8][9][10].Linear segment detection methods are another prevalent method in pattern recognition category. These methods are based on the assumption that vessels are piecewise linear segments and profile of the vessel cross section can be approximated by a Gaussian-shape. Some match filters which properly describe these linear segments were designed and applied [11][12][13][14]. If a vessel width matches with the scale of filter, there will be a strong response. But they also respond to non-vessel samples such as Bright lesion and optic ...

A Region Growing Vessel Segmentation Algorithm Based on Spectrum Information

“…Generally, the seed points should be in the region of interest, while trying to avoid the edge of the region. To achieve this, first we use Sobel [ 12 ] operator to extract the edge information, that combined which and the obtained primary feature direction. We cut down in the direction that perpendiculars to the vascular structure's feature direction.…”

Global Transformer and Dual Local Attention Network via Deep-Shallow Hierarchical Feature Fusion for Retinal Vessel Segmentation

“…, (6) where dilated conv r (•) denotes the dilated convolution with a dilated rate of 𝑟. Additionally, we design an edge detection block to extract the blood vessel details in the encoder features. The multi-directional Sobel operator [44,45] is adopted to detect gradient changes between retinal blood vessel pixels and background pixels. The output of the edge detection block can be calculated by: 𝐺 = √𝐺 0 °2 + 𝐺 45 °2 + 𝐺 90 °2 + 𝐺 135 °2 (7) where…”

Automatic Segmentation and Detection System for Varicocele Using Ultrasound Images

“…Ghadiri et al [23] devised a fuzzy inference and edge detection method for identifying vessels in retinal analysis to extract information such as blood vessel alterations. They utilized linear structures and split the image into overlapping windows.…”

“…This condition will also exclude any circular shapes identified in the previous step if these shapes are too large to be veins or too small to be identified as dilated (varicocele) veins. The lower bound (23) was experimentally determined to reduce false positive results. The upper bound (60) was based on medical information about varicocele.…”