ArcMap Image Classification Tutorial : 2.3- The spatial filters

 The spatial filters represent another method of digital processing used for the enhancement of an image. These filters are designed to bring out or remove specific features of an image based on their spatial frequency. The spatial frequency is related to the concept of texture. It refers to the frequency of variation of the different tones that appear in an image. The regions of an image where the texture is “rough” are the regions where the changes in different shades are steep; these regions have a high Spatial frequency. The “smooth”regions have a variation of tones that is more gradual over several pixels;these regions have a weak Spatial frequency Space. The spatial filtering method implies moving a “window” the size of a few pixels (eg. 3 of 3, 5 out of 5,etc.) above each image pixel. Then, we apply a mathematical processing using the values of the pixels under the window and replace the value of the central pixel by the result obtained. The window is moved along the columns and lines of the image, one pixel at a time, repeating the calculation until the whole image has been filtered. By modifying the calculation performed within the window, is possible to enhance or remove different types of features present in an image.



A low-pass filter is designed in  order to put in evidence the regions large and homogeneous enough with pixels of similar intensity. This filter reduces the smaller details of an image. Therefore,it is used to straighten an image. The average and median filters, often used with radar images, are low-pass filter examples. The high-pass filters are the opposite: they are used to enhance the little details of an image. A high-pass filter can be defined by applying firstly a low-pass filter to an image and then subtract the result from the original image, producing a new image where details having a high Spatial frequency are enhanced. The directional filters or the filters that detect contours are used to enhance the linear features of an image such as roads or field boundaries. These filters can, also, be designed to enhance the characteristics that have a certain orientation in the image.

The enhancement of images is, basically, anything that facilitates the visual interpretation of an image. In certain cases ( as low-pass filters), the result can result disappointing but it empowers the interpreter to discern the low frequency spatial elements among the clutter of the high frequencies of the image . The enhancements are, often, applied for specific reasons. Therefore,for a given image, if dealing with different applications, strong differences are possible.

Unlike enhancement, filtering can take place in two different ways: by modifying the visualization of the image or the values of the image. If you decide to apply a definitely filtering when working on the classification of the modified image, you will have to create a new image at the end of the filtering process .

Filtering with ArcMap You will find a tool in the toolbox to perform the filtering of images : Spatial Analyst Tools->Neighborhood->Filter .   

But this option is very limited and requires the creation of a new image.

The most appropriate ArcMap tool to work with filtering is the Image Analysis extension.

Image Analysis To open the window of Image Analysis, go in the Windows Menu -> Image Analysis

Then,
the  Image Analysis window appears   

The interface has five components: the list of layers, the Options panel,
the Display panel, the Processing panel and the Measurement panel.

List of layers

The list of layers reflects all the raster layers that are found in the active data block, including the picture layer of the mosaic, the image service and WCS layers. The image catalogue layer is not visible and cannot be used in the Image Analysis window. The list of layers does not replace the table of contents, but it allows you access to the Layer Properties dialog box and to delete layers. It allows you, also, to accelerate the layer. An accelerated raster layer appears with this icon  

To rearrange the layers, you must use the table of contents.

Options

The Analysis Options button image is at the top of the Image Analysis window. It allows opening the Image Analysis Options dialog box where you can set default values for some of the window tools, such as the definition of the red channels and near infrared to use with the NDVI tool, the altitude and azimuth values to use with the Shaded Relief tool and the standard deviation value to be used with the Standard deviation stretching method. You must change the options before to use the tool.

Viewing

The Viewing tools allow improving the data appearance. This component includes contrast cursors, brightness, gamma correction and transparency, a check box to use the
dynamic adjustment of the beach , and the tools Sweep and Flicker tools to compare two layers of superimposed data.

Processing

 rectification, the convolution filters, and mosaicking.

New layers are added to the list of layers and the table of contents when you use the Treatment panel tool. Indeed, all processing tools generate a new temporary raster layer that uses functions to process the data.The functions allow a quick processing application instead of creating another data file (whose generation can take some time) where the process is applied permanently.To save the layer, you can export it into a new raster data set or save the layer file. If you close ArcMap without saving, you lose this raster layer.

Measurement

The Measurement option includes tools for measuring a point, a distance,an angle, a height, a perimeter and a surface of an Image (raster or mosaic data set)with information from thesensor ( or geodata transformation). The height can be measured along an object, its shade, or both. You can also use a DTM to perform a measure along a surface rather than the projected area

Available filters with Image Analysis

To set the filter to be used and applied it to your image:

1- Select in the list of Image Analysis layers the image to filter

2- In the Processing panel select the type of filter to be used in the filter list available

3- Click the Filter button to create the filtered temporary layer

The available filters relate to three main types of filters :

The low pass filters (Smoothing)

Smoothing filters (low – pass) straighten data by reducing local variations and removing the noise. The low pass filter calculates the average value for each neighbouring pixel. The result is that the average of the high and low values of each neighbour will be reduced, which will reduce the data extreme values.You have two filters of this type in the drop-down menu:

  • Smooth Arithmetic Mean
  • Smoothing 5 × 5

The high pass filters (Sharpening)

The Sharpness filter accentuates the values comparative difference among neighbours. A high pass filter calculates the sum of Statistics focal length for each cell of the input using a weighted neighbourhood of the core. It highlights the boundaries among features (for example, when a body of water meets the forest), accentuating, thus, the contours among the objects. The high pass filter is called contour improvement filter. The core high pass filter identifies which cells to use in the neighbourhood and their weights. You have this type of filters in the drop-down menu :

  • Sharpen
  • Sharpen more
  • Sharpening 3 × 3
  • Sharpening 5 × 5

Edge detecting filters

The rest of the filters relates to the detection of edges of geographical objects.
Gradient filters can be used for edge detection in 45 degree increments(Gradient North, Gradient North-west …).

The Laplacian filters are, often, used for contour detection. It is,
usually, applied to an image that first is smoothed to reduce its sensitivity noise
(Laplacian 3 × 3, Laplacian 5 × 5).

Line detection filters such as gradient filters, can be used to perform an edge detection (horizontal line, vertical line …). You will get better results if you apply a smoothing algorithm before an edge detection algorithm.
The Sobel filters are, as well, used for edge detection.

Leave a Reply

Your email address will not be published. Required fields are marked *