ILMSimage 2.4 Cell Creation
Overview • Project Settings • Cell Creation • Feature Calculation • Classification • Miscellaneous
Contents |
Cell Creation
The cell index summarizes areas with similar pixel features as "cells" in the image. Image analysis with ILMS depends completely on cells. The input panel Cells is used to control cell size and the distribution of cells with three input parameters. External Borders will be influenced by newly created cells.
Main page of ILMSImage tutorial.
Parameters
- Minimum Modulation for input A suitable value for the input in Modulation (Mean Cell Size) is difficult to guess, so the button Determine starts a routine that calculates the mean modulation of the selected images. The result is shown in the box left of the button. The entry in Modulation (mean Cell Size) should be considerably higher than the result in Minimum Modulation for Input.
- Modulation (Mean Cell Size) is the most significant input parameter. The input determines the size of all of the resulting homogeneous areas referred as "cells" resulting from the cell creation procedure. Smaller values of modulation lead to smaller cells. Input values from 0 to 1 are defined.
- Homogeneous Cell Size determines the distribution of different cell sizes after the cell creation process. Small values ( 0.0 - 0.1) lead to a "natural" size distribution with highly different areas covered by a single cell. This is the normal case, but it can be useful to keep the cell size rather equal in order to carry out unusual classification tasks.
- Minimum Cell Size assures that all cells consists at least of the amount of pixels in the entry. If cells are smaller than the entry, ILMSimage distributes their pixels equally among neighboring cells. This process is independent from image data.
- Provide External Borders transfers all borders from the given shape to the cells. ILMSimage may add additional borders within the predestinated cells. The entry of External Borders is optional.
ILMSImage uses a tile concept to process large data layers efficiently. The maximum size of the applied tiles can be restricted by an input in Tilesize [MB] at the lower left corner of the ILMSimage menu. Under normal conditions ILMSimage uses up to 10 tiles simultaneously, so for 1000 MB free working memory an entry of "100 MB" is recommended.
Accomplish Cell Creation
After clicking the [Run] button the selected operations are carried out and a small window with a progress bar will show up and provide information about the cell generation status.
This can take some time according to the size of the test area and the selected parameters.
After the cell creation process has finished QuantumGIS shows the resulting cell borders with a semitransparent raster GIS layer. Raster layers can be visualized considerably quicker than vector layers with many polygons. If cell creation has finished successfully, ILMSimage will show the feature calculation panel.
If cell borders should be represented by shapes, the ILMSimage panel [Export] supplies tools to convert the rasterized cell index into a shape layer and show it on the QuantumGIS canvas. It is possible to smoothen the generated polygons adaptively which can be done by activating the check box Smooth arcs [grade] and Optimize Vertex Density. The degree of smoothing can be determined in the input box where polygons are more strongly smoothed with increasing degree.
Background
The derivation of statistically homogeneous image areas is a key process in object-based image analysis. It is based on the assumption that the data material of modern remote sensing sensors, which is getting more complex due to increasing data density, cannot be analyzed adequately because pixels from common image analysis are used as a reference. As an alternative, object-based image analysis offers the creation of new areas as a basis for subsequent analysis. This process is often called segmentation and, in the context of ILMSImage as cell creation. It summarizes neighboring pixels on the basis of statistical similarities to areas (cells) which are as homogeneous as possible. This process does not only represent reduction of complexity of extensive input data but it also offers a new perspective on those data by providing new features (which are only rarely used in image analysis) for subsequent analysis. One of those features is, for example, the form of a cell or its spatial context which contains, among other things, the relation to the neighboring cells. Features for form and context are perceived as subordinate features in pixel-based analysis of remote sensing data due to the uniformity of the area, mostly the rectangular pixel.
For the segmentation process, i.e. the derivation of such reference areas from images, various suggestions and methods are provided in the corresponding literature. Some of them are summarized in Haralick & Shapiro (1985), Pal & Pal (1993) and Freixenet et al. (2002). For a basic understanding of ILMSImage it is sufficient to know that the cell creation process used in the software is based on a combination of the methods Region Growing and Watershed Analysis. It was developed with the aim to minimize the degrees of freedom which are available to the user for adaption. For an exact application of the method more detailed options are available. The result of the cell creation using ILMSImage is a first abstraction of the original image content, all other analyses are carried out using this cell image.
Modulation
The term Modulation derives from the maximum contrast between neighboring pixels in the image data (Schowengerdt, 2007). The parameter of modulation controls the average size of emerging cells. The following relation has to be considered: with increasing modulation the average size of cells increases, their number decreases accordingly, with decreasing modulation the average size of cells decreases but their number increases.
The range of values of the parameter lies between 0 and 1; it has to be noted that the extreme values do not lead to meaningful results under any circumstances. The value 0.03 is a predefined modulation value which has proven to be a good starting point for cell creation for a combination of spatially more highly resolved panchromatic data and spatially more lowly resolved multi-spectral data (as they are used in the current example). This default setting can be modified by the user concerning the cell creation parameter, so that it corresponds to the typical area size which is required for the analysis. The result of a new cell creation is a raster data set which has the pixel size of the most highly resolved input data, which can be found in the project directory and ends in _index. As ILMSImage requires exclusive access to the corresponding file for more operations, it is not automatically loaded into the map view. For visualizing the cell creation result the option of Export as Shapefile (see below) is more suitable.
Overview • Project Settings • Cell Creation • Feature Calculation • Classification • Miscellaneous
