China
Africa
Explore chapters and articles related to this topic
Feature Extraction and Classification for Environmental Remote Sensing
Published in Ni-Bin Chang, Kaixu Bai, Multisensor Data Fusion and Machine Learning for Environmental Remote Sensing, 2018
In spatial feature extraction, region-growing methods are commonly used to segment images for feature extraction by considering spatial relationships among pixels, based on the assumption that pixels nearby (or neighboring pixels) often have similar data values or attributes. Therefore, the common practice is to find a data center and then compare this central pixel with its neighbors to examine their similarities. For example, spectrally similar and spatially closing pixels can be gathered to form one specific cluster. In spatial feature extraction methods, connectivity and similarity are two basic aspects commonly applied for feature extraction. Connectivity is defined in terms of pixel neighborhoods while similarity refers to texture property in either grey level or shape. A typical method is the statistical region merging, which begins by building the graph of pixels using 4-connectedness with edges weighted by the absolute value of the intensity difference and then sorting those edges in a priority queue and merging the current regions belonging to the edge pixels based on certain criteria (Boltz, 2004).
Saliency-based segmentation of dermoscopic images using colour information
Published in Computer Methods in Biomechanics and Biomedical Engineering: Imaging & Visualization, 2022
The region-based methods group similar neighbouring pixels into larger regions according to a given criterion; see, e.g. the classical methods (Iyatomi et al. 2008; Celebi et al. 2008). The method proposed in (Iyatomi et al. 2008) consists of four phases: 1. initial tumour area detection by two filtering operations before the selection of a threshold; 2. regionalisation by merging small isolated regions created in the first phase; 3. tumour area selection by selecting appropriate areas according to predefined criteria; (4) region growing. In (Celebi et al. 2008), after a pre-processing step including the smoothing of the image, a fast and unsupervised technique is applied based on the statistical region merging. Next, to eliminate spurious detected regions belonging to the background, a post-processing step is executed. Typically, the region-based methods distinguish the lesion components (regions) by standard image processing techniques such as, for instance, statistical region merging (Celebi et al. 2008), modified JSEG (Celebi et al. 2007), watershed (Wang et al. 2011), and complex networks (Aksac et al. 2017). Thus, the identification problems of the region-based methods are directly connected to the kind of employed processing technique.