China
Africa
Explore chapters and articles related to this topic
A Genetic Algorithm for Image Segmentation
Published in Takushi Tanaka, Setsuo Ohsuga, Moonis Ali, Industrial and Engineering Applications of Artificial Intelligence and Expert Systems, 2022
A. Calle, W.D. Potter, S.M. Bhandarkar
Once an atomic pixel has passed the initial test, we apply a region-growing technique that ensures the compactness feature of an atomic cluster. In our case, we have focused on preventing the region-growing algorithm from straying into “rough” areas i.e. areas in the image with lot of gray level detail. The region-growing procedure is as follows: We treat each of the atomic pixels as the first member of an atomic cluster and incorporate the adjacent pixels whenever they satisfy the region-growing test. This test is positive if the variance in the local neighborhood of the pixel (where the local neighborhood is window of size 3x3 centered at the atomic pixel, for instance) is lower than some predefined threshold. The process is repeated by enlarging the search window around the atomic pixel. The process is halted when we reach a window in which it is not possible to label a new pixel as belonging to the atomic cluster. The region-growing procedure outlined above ensures the connectivity of the atomic clusters.
Basics of Image Processing
Published in Maheshkumar H. Kolekar, Intelligent Video Surveillance Systems, 2018
In noisy images, edge detection is sometimes difficult. In such cases, the region-based approach is preferred. A region is a group of pixels with similar properties. Region growing is the simplest image segmentation approach, that groups pixels or sub-regions into larger regions based on some pre-defined criteria. Initially, one set of seed points are selected. Based on some criteria, all neighbors of seed points are tested. If neighbors are similar to the seed point based on the criteria, neighbors will be merged into the region. This procedure is repeated until the growth of the region is stopped. The similarity measure can be selected as gray level, texture, color, or shape. The selection of seed points, and the similarity criteria, are based on the problem under consideration.
Computer Analysis of Mammograms
Published in Paolo Russo, Handbook of X-ray Imaging, 2017
Chisako Muramatsu, Hiroshi Fujita
There have been many studies investigating the computerized schemes for mass segmentation on mammograms. Oliver et al. (2010) provided a comprehensive review on such algorithms. Advantages and disadvantages of different segmentation methods are described in Cheng et al. (2006). In their review, the segmentation techniques are grouped into four types, that is, region-based methods, contour-based methods, clustering methods and model-based methods. Typical region-based techniques include region growing methods (Zucker 1976) and watershed methods (Beucher and Lenteuejoul 1979). The region growing method is a classic algorithm which expands a seed region by including the neighbor pixels with a specific criterion (generally pixel values). In region growing methods for mass segmentation, gradient or edge information was often used as an additional stopping criterion (Barman and Granlund 1994; Kupinski and Giger 1998; Petric et al. 1999). In some studies, a probabilistic model approach was also incorporated (Kupinski and Giger 1998; Marti et al. 2003; Kinnard et al. 2004). Other groups investigated improving the seed point placement (Zheng et al. 1995; Qi and Snyder 1998; Zhang et al. 2004).
Automatic liver tumour segmentation in CT combining FCN and NMF-based deformable model
Published in Computer Methods in Biomechanics and Biomedical Engineering: Imaging & Visualization, 2020
Shenhai Zheng, Bin Fang, Laquan Li, Mingqi Gao, Yi Wang, Kaiyi Peng
Until now, many methods have been developed for segmenting liver tumours (Luo et al. 2014), but only few methods achieve acceptable results. These methods include the thresholding (Moltz et al. 2008; Wels et al. 2008), graph cuts (Wels et al. 2008), watershed (Ray et al. 2008), region growing (Wong et al. 2008; Baâzaoui et al. 2017), classification (Zhang et al. 2011; Boas et al. 2015; Sajith and Hariharan 2015), statistical analysis (Kumar et al. 2016) and so on. Based on thresholding and morphological processing, a semi-automatic segmentation method of liver tumours was achieved by Moltz et al. (2008). The proposed discriminative model-constrained graph cuts approach (Wels et al. 2008) was applied to the challenging task of detection and delineation of pediatric brain tumours. An interactive method combining graph cuts and watershed was used to delineate tumours in 3D CT images (Stawiaski et al. 2008). Region growing methods for tumour segmentation were studied by selecting a seed point and expanding it with some specified knowledge constraints (Wong et al. 2008; Baâzaoui et al. 2017). Support vector machines (SVM) classification (Zhang et al. 2011) and statistical analysis (Kumar et al. 2016) were also researched to extract tumours from CT images. Usually, these methods are just based on the pixel or voxel intensity. However, because of the noise, intensity similarity and heterogeneity, these methods cause boundary leakage, under-segmentation or over-segmentation results easily.
Image-based finite-element modeling of the human femur
Published in Computer Methods in Biomechanics and Biomedical Engineering, 2020
Cristina Falcinelli, Cari Whyne
To reduce the manual effort required, differing levels of automation have been developed for segmentation of bones from CT images. These approaches include intensity-based thresholding, edge-based detection and region growing (Table 5). Thresholding-based methods utilize an intensity value (global, local or adaptive) to separate bone from other tissues; it is often considered the first stage of segmentation followed by manual or automated refinement methods. Global thresholding works well on images characterized by a bimodal intensity distribution, but the majority of diagnostic images are not characterized by a bimodal intensity distribution. To overcome this issue, adaptive thresholding has been adopted that allows to divide the image into multiple sub-images and apply a different threshold to each sub-image. However, how to divide the image and estimate thresholds are still open issues. Edge-based algorithms utilize the local change of intensity in small regions to segment bone tissue (Rathnayaka et al. 2011); while they are characterized by high repeatability, edge-based methods are highly susceptible to artifacts and noise. Region growing utilizes seed points (considered as representative pixels) to extract all similar connected pixels with a region; this approach is fast, but it can be highly dependent on the choice of the seed points and noise can lead to holes and disconnected boundaries in the segmented region. Similar to thresholding-based methods, region growing is often used as one step of multiple segmentation operations.
Spatial search and a three level model based water layer extraction from C-band SAR image
Published in Annals of GIS, 2021
C. Bipin, C. V. Rao, P. V. Sridevi
Object Based Image Analysis (OBIA) brought in a paradigm shift from pixel to segment-based image processing (Blaschke 2010). It is extensively used in many remote sensing applications, mostly in multispectral, hyper spectral and thermal remote sensing (Bechtel, Ringeler, and Bohner 2008). Region Growing algorithms are widely used for image segmentation. These methods segment images based on a similarity condition to properly find the boundary between two segments. These algorithms are tailored to work under varying lighting conditions, which is common in optical images.