Stay organized with collectionsSave and categorize content based on your preferences.
Performs G-Means clustering on the input image. Iteratively applies k-means followed by a normality test to automatically determine the number of clusters to use. The output contains a 'clusters' band containing the integer ID of the cluster that each pixel belongs to. The algorithm can work either on a fixed grid of non-overlapping cells (gridSize, which can be smaller than a tile) or on tiles with overlap (neighborhoodSize). The default is to use tiles with no overlap. Clusters in one cell or tile are unrelated to clusters in another. Any cluster that spans a cell or tile boundary may receive two different labels in the two halves. Any input pixels with partial masks are fully masked in the output. This algorithm is only expected to perform well for images with a narrow dynamic range (i.e., bytes or shorts).
See: G. Hamerly and C. Elkan. 'Learning the k in k-means'. NIPS, 2003.
Neighborhood size. The amount to extend each tile (overlap) when computing the clusters. This option is mutually exclusive with gridSize.
gridSize
Integer, default: null
Grid cell-size. If greater than 0, kMeans will be run independently on cells of this size. This has the effect of limiting the size of any cluster to be gridSize or smaller. This option is mutually exclusive with neighborhoodSize.
uniqueLabels
Boolean, default: true
If true, clusters are assigned unique IDs. Otherwise, they repeat per tile or grid cell.
[[["Easy to understand","easyToUnderstand","thumb-up"],["Solved my problem","solvedMyProblem","thumb-up"],["Other","otherUp","thumb-up"]],[["Missing the information I need","missingTheInformationINeed","thumb-down"],["Too complicated / too many steps","tooComplicatedTooManySteps","thumb-down"],["Out of date","outOfDate","thumb-down"],["Samples / code issue","samplesCodeIssue","thumb-down"],["Other","otherDown","thumb-down"]],["Last updated 2024-07-13 UTC."],[[["\u003cp\u003eThe GMeans algorithm automatically determines the optimal number of clusters for image segmentation using iterative k-means and a normality test.\u003c/p\u003e\n"],["\u003cp\u003eIt outputs an image with a 'clusters' band, assigning each pixel an integer ID corresponding to its cluster.\u003c/p\u003e\n"],["\u003cp\u003eUsers can control segmentation granularity through \u003ccode\u003egridSize\u003c/code\u003e for non-overlapping cells or \u003ccode\u003eneighborhoodSize\u003c/code\u003e for overlapping tiles.\u003c/p\u003e\n"],["\u003cp\u003eClusters are independent within each cell or tile, potentially leading to different labels for the same cluster across boundaries.\u003c/p\u003e\n"],["\u003cp\u003eThe algorithm is best suited for images with a narrow dynamic range, like those with byte or short data types.\u003c/p\u003e\n"]]],["The G-Means algorithm performs image clustering by iteratively applying k-means and a normality test to determine the optimal number of clusters. It outputs an image with a 'clusters' band, assigning each pixel to a cluster. It can operate on a fixed grid (gridSize) or tiles with overlap (neighborhoodSize), with default being tiles without overlap. Input images should have a narrow dynamic range and pixels with partial mask will be fully masked in the output. Clusters can be assigned unique ID's or repeat per tile.\n"],null,["# ee.Algorithms.Image.Segmentation.GMeans\n\nPerforms G-Means clustering on the input image. Iteratively applies k-means followed by a normality test to automatically determine the number of clusters to use. The output contains a 'clusters' band containing the integer ID of the cluster that each pixel belongs to. The algorithm can work either on a fixed grid of non-overlapping cells (gridSize, which can be smaller than a tile) or on tiles with overlap (neighborhoodSize). The default is to use tiles with no overlap. Clusters in one cell or tile are unrelated to clusters in another. Any cluster that spans a cell or tile boundary may receive two different labels in the two halves. Any input pixels with partial masks are fully masked in the output. This algorithm is only expected to perform well for images with a narrow dynamic range (i.e., bytes or shorts).\n\n\u003cbr /\u003e\n\nSee: G. Hamerly and C. Elkan. 'Learning the k in k-means'. NIPS, 2003.\n\n| Usage | Returns |\n|-------------------------------------------------------------------------------------------------------------------------------------------|---------|\n| `ee.Algorithms.Image.Segmentation.GMeans(image, `*numIterations* `, `*pValue* `, `*neighborhoodSize* `, `*gridSize* `, `*uniqueLabels*`)` | Image |\n\n| Argument | Type | Details |\n|--------------------|------------------------|----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|\n| `image` | Image | The input image for clustering. |\n| `numIterations` | Integer, default: 10 | Number of iterations. Default 10. |\n| `pValue` | Float, default: 50 | Significance level for normality test. |\n| `neighborhoodSize` | Integer, default: 0 | Neighborhood size. The amount to extend each tile (overlap) when computing the clusters. This option is mutually exclusive with gridSize. |\n| `gridSize` | Integer, default: null | Grid cell-size. If greater than 0, kMeans will be run independently on cells of this size. This has the effect of limiting the size of any cluster to be gridSize or smaller. This option is mutually exclusive with neighborhoodSize. |\n| `uniqueLabels` | Boolean, default: true | If true, clusters are assigned unique IDs. Otherwise, they repeat per tile or grid cell. |"]]
