China
Africa
Explore chapters and articles related to this topic
Research on insect pest image detection and recognition based on bio-inspired methods
Published in Limiao Deng, Cognitive and Neural Modelling for Visual Information Representation and Memorization, 2022
Thus the saliency map of the input image can be obtained. To facilitate the subsequent processing, the saliency map was normalized to the range of 0 to 1. Then, the object was separated from the background using an adaptive threshold that was computed using Otsu's method. If the saliency of a pixel is larger than the threshold, the pixel value was set to 1; otherwise, it was set to 0. This would result in a binary image, on which the largest connected region usually corresponded to the object of interest. Then we chose the minimum bounding rectangle of the largest connected region as the ROI. Fig. 8.4 demonstrates the saliency maps and the ROIs of some image samples. The first row shows raw images, the second row their corresponding saliency maps, and the third row the corresponding ROIs. Though there was great variation in the colour and backgrounds of the sample images, our method could accurately target the object and extract the ROIs from the complex backgrounds.
A Survey of Uncertain Data Clustering Algorithms
Published in Charu C. Aggarwal, Chandan K. Reddy, Data Clustering, 2018
The work in [48] improves on the work of [16] and designs a pruned version of the UK-means algorithm. The idea here is to use branch-and-bound techniques in order to minimize the number of expected distance computations between data points and cluster representatives. The broad idea is that once an upper bound on the minimum distance of a particular data point to some cluster representative has been quantified, it is necessary to perform the computation between this point and another cluster representative, if it can be proved that the corresponding distance is greater than this bound. In order to compute the bounds, the minimum bounding rectangle for the representative point for a cluster region is computed. The uncertain data point also represents a region over which the object may be distributed. For each representative cluster, its minimum bounding rectangle (MBR) is used to compute the following two quantities with respect to the uncertain data point: The minimum limit on the expected distance between the MBR of the representative point and the uncertain region for the data point itself.The maximum limit on the expected distance between the MBR of the representative point and the uncertain region for the data point itself.
Building and Road Extraction from LiDAR Data
Published in Jie Shan, Charles K. Toth, Topographic Laser Ranging and Scanning, 2018
Franz Rottensteiner, Simon Clode
Car parks are not considered to be roads. However, as car parks and roads have similar surface and reflectance properties, it is difficult to detect and eliminate car parks. By defining a maximum acceptable road width prior to processing, very wide unconnected car parks can be removed from the binary image. As roads form a network of long, thin connected objects, the area ratio of each individual road segment and the corresponding minimum bounding rectangle will decrease as the length of the smallest side in the minimum bounding rectangle increases. Large isolated blobs can be detected in the image using this ratio, thus allowing the removal of any unconnected car parks from the final binary image of road pixels.
Identifying China’s polycentric cities and evaluating the urban centre development level using Luojia-1A night-time light data
Published in Annals of GIS, 2022
Zhiwei Yang, Yingbiao Chen, Zihao Zheng, Zhifeng Wu
The elongation () represents the spatial distribution of the NC in different directions. and represent the lengths of the major and minor axes, respectively, of the minimum bounding rectangle of the NC. In the process of deriving NCs, we found that some NTL pixels with high values (location in the same area as a narrow road) can form an NC. Therefore, in identifying urban centres, we need to eliminate those NCs that represent narrow roads. The minimum bounding rectangle of an NC can help us distinguish whether an NC is long and narrow. In general, if of an NC is equal to or greater than 5, then the NC is not an urban centre.
Efficiently identifying closed roads by integrating and indexing open data
Published in Journal of the Chinese Institute of Engineers, 2021
Ya-Hui Chang, Shu-Han He, Chih-Wei Tseng
The spatial data processed in this research are the set of closed roads. Based on the output of the preprocess algorithms, each closed road is represented as a point sequence, and is like a line with an arbitrary shape. To reduce the load of directly processing each point, we adopt the minimum bounding rectangle (MBR) of the line to build an R-tree index first. A sample road network is shown in Figure 4(a), where each Ei represents a closed road and the gray area MEi is its MBR. Several adjacent MBR’s will be combined to form a bigger MBR, which is represented in the upper level of the R-tree according to the construction algorithm. For example, the MBR M1 is formed based on the MBR’s of the closed roads E1, E2, and E3. By applying the above procedure recursively, we can construct the corresponding R-tree index, as shown in Figure 4(b). Here we assume that the order of the tree, i.e., the maximum number of values which can be represented in a node, is three. Besides, in the leaf nodes, we represent the pair of the event and its corresponding MBR, such as (E1, ME1), so that the whole event can be returned to the user and provide detailed information.
Collision detection during planning for sheet metal bending by bounding volume hierarchy approaches
Published in International Journal of Computer Integrated Manufacturing, 2018
D. Raj Prasanth, M. S. Shunmugam
Unfortunately, the results from this method are also very similar to Figure 8(a) and therefore are not satisfactory. The eigenvectors of the covariance matrix are not the only way to determine the principal directions of the oriented bounding box. O’Rourke (1985) showed that it is possible to compute the minimum volume oriented bounding box from the convex hull of the model. He showed that the minimal bounding box needs to have at least two of its faces touching two edges of the convex hull. His work was based on the earlier works of Freeman and Shapira (1975) and Toussaint (1983) who showed that in 2D space the minimum bounding rectangle has one of its edges coincide with the edge of the 2D convex hull. However, although the 2D minimum orientated rectangle problem was solved by Toussaint in O(n) times, O’Rourke’s algorithm for 3D space has O(n3) complexity and is therefore not useful for practical purposes. A more practical solution is offered by Barequet et al. (1996), wherein at first the eigenvectors are calculated using the covariance method. The eigenvector corresponding to either the longest or shortest eigenvalue is taken as the first axis of the OBB. All points of the model are then projected onto a plane perpendicular to this axis. The convex hull for these points is computed, and their minimum bounding rectangle is determined. The axes of the minimum bounding rectangle provide the second and third axes of the OBB. Their experimental results show that taking the eigenvector corresponding to the smallest eigenvalue as the first axis provides the best fit OBB.