areaDetector Plugin NDPluginProcess

Contents

• Overview
• Configuration
• Screen shots

Overview

NDPluginProcess performs arithmetic processing on NDArray data. It performs the following operations in the order listed. Each of these operations can be individually enabled and disabled.

1. Subtracts a background array which has been previously acquired.
2. Divides by a flat field array which has been previously acquired, and then multiplies by a flat field scale factor.
3. Multiplies by a scale factor and adds an offset.
4. Clips to a maximum specified value.
5. Clips to a minimum specified value.
6. Applies a recursive digital filter.
7. Converts to the specified output data type.
8. Exports the processed data as a new NDArray object.

If any of the above operations is enabled, then the array is first converted to NDFloat64 data type, i.e. double-precision float. The operations are all performed in double-precision, and then the array is converted to the specified output data type.

NDPluginProcess is both a recipient of callbacks and a source of NDArray callbacks. This means that other plugins, such the NDPluginStdArrays, NDPluginStats, and NDPluginFile plugins can be connected to an NDPluginProcess plugin, in which case they will use the processed data.

NDPluginProcess is fully N-dimensional. It can be used for 2-D images, 3-D (color) images, or any type of N-dimensional data.

NDPluginProcess inherits from NDPluginDriver. The NDPluginProcess class documentation describes this class in detail.

NDPluginProcess.h defines the following parameters. It also implements all of the standard plugin parameters from NDPluginDriver. The EPICS database NDProcess.template provide access to these parameters, listed in the following table. Note that to reduce the width of this table the parameter index variable names have been split into 2 lines, but these are just a single name, for example NDPluginProcessSaveBackground.

Recursive filter implementation

The recursive filter performs filtering in the time (not spatial) domain. It is implemented in a fairly general manner, so that a variety of filters can be implemented. These include integrating filters and differentiating filter types. The recursive filter stores one "filter" array internally. Using this internal filter array, and a new input array, it computes an output array, and a new version of the filter array. The equations governing the output array and new filter array are:

O[n] = OOffset + OScale*((OC1 + OC2/N)*F[n-1] +
                         (OC3 + OC4/N)*I[n])
F[n] = FOffset + FScale*((FC1 + FC2/N)*F[n-1] +
                         (FC3 + FC4/N)*I[n])
On filter reset
F[0] = ROffset + RC1*F[0] + RC2*I[0] 
  
where
  I[n] = New input array from callback
  I[0] = First input array after a filter reset
F[n-1] = Stored filter array
  F[n] = New filter array
  F[0] = Current filter array when filter is reset.  
         May be a copy of I[0] if there was no valid filter array.
     N = Current value of NumFiltered
  O[n] = Output array passed to clients

Predefined filters

The NDProcess.template database implements the following predefined filters using the $(P)$(R)FilterType record. The implementation of these predefined filter types is done entirely in the database, not in the plugin code. The database simply downloads values of OC1-OC4, FC1-FC4, and RC1-RC2 that result in predefined filter behaviours. The operation of the recursive filter is by no means limited to these fixed filter types, they are simply provided as a convenience, and can be easily modified or extended. Note that the database does not download values of OOffset, OScale, FOffset, FScale, or ROffset, so these remain unchanged when a new filter is selected. The equations below do not include these offset and scale values, but they are useful in many circumstances.

The NDProcess_settings.req file for save/restore saves the values of all of the filter coefficients, and downloads them when EPICS initialized at iocInit. It also saves the FilterType, but does not process this record at iocInit, so that any customized filter coefficients will be preserved, and will not be replaced by the defaults for that filter type.

Recursive Average

The recursive average filter does input array averaging. It is defined as:

O[n] = F[n] = (1-1/N)*F[n-1] + 1/N*I[n]

N is the characteristic number of arrays in the average. For example, if N=100, then each new array is weighted by 0.01 and the previous filter value is weighted by 0.99. This is an infinite impulse response (IIR) filter, because the effect of one array never complelely dissappears, but rather decays exponentially. When this filter is reset the filter array is initialized with the first input array. This filter can be used to decrease the noise and increase the dynamic range of a repetitive input signal with a small signal. In that case it can be useful to use set EnableOffsetScale=Enable and set Scale to a large enough number to more fully use the dynamic range of the output data type. The averaged signal will then better use the dynamic range of the output data type.

Average

The average filter does input array averaging. It is defined as:

O[n] = F[n] = F[n-1] + 1/N*I[n]

N is the number of arrays in the average. For example, if N=100, then each new array is weighted by 0.01. When this filter is reset the filter array is initialized to 0. This filter typically cannot be run forever, because the output grows monotonically and will lead to overflow. However, if AutoResetFilter is enabled then the filter will be reset when NumFiltered=NumFilter. If ArrayCallbacks is "Array N only" then callbacks will be done only with the final averaged value.

Sum

The sum filter does input array summing. It is defined as:

O[n] = F[n] = F[n-1] + I[n]

This filter simply computes the sum of input arrays. When this filter is reset the filter array is initialized to 0. It is often necessary to set the output data type to one with a larger maximum value than the input array to prevent overflow. If AutoResetFilter is enabled then the filter will be reset when NumFiltered=NumFilter. If ArrayCallbacks is "Array N only" then callbacks will be done only with the final summed value.

Difference

The difference filter computes the difference between frame N and frame N-1. It is defined as:

O[n] = -F[n-1] + I[n]
F[n] = I[n]

Note that the difference will often be a small signed number. If the output datatype is unsigned, or if a display client expects unsigned numbers, then it can be useful to set OOffset to a non-zero value to add an offset to the result. For example, if the output data type is UInt8 (unsigned 8-bit), then setting OOffset to 128 will produce an unsigned image centered at 128. Similarly OScale can be used to increase the range of the difference by multiplying by a factor greater than 1, although this will not increase the dynamic range of the result, only the absolute range.

Recursive Average Difference

This filter computes the difference between frame N and the recursive average of previous frames. It is a combination of the Difference and Recursive Average filters described above. The new filter is a recursive average, but the output is the difference between the current frame and that recursive average. It is defined as:

O[n] = -F[n-1] + I[n]
F[n] = (1-1/N)*F[n-1] + 1/N*I[n]

N is again the characteristic number of arrays in the average.

Copy to Filter

This filter simply copies the input array to the filter array and the output array. It is defined as:

O[n] = F[n] = I[n]

This filter can be used to load the filter (F) with a certain array, which is then subsequently used for some other filter type.

Configuration

The NDPluginProcess plugin is created with the NDProcessConfigure command, either from C/C++ or from the EPICS IOC shell.

NDProcessConfigure(const char *portName, int queueSize, int blockingCallbacks,
                   const char *NDArrayPort, int NDArrayAddr,
                   int maxBuffers, size_t maxMemory,
                   int priority, int stackSize)
  

For details on the meaning of the parameters to this function refer to the detailed documentation on the NDProcessConfigure function in the NDPluginProcess.cpp documentation and in the documentation for the constructor for the NDPluginProcess class.

Screen shots

The following is the MEDM screen that provides access to the parameters in NDPluginDriver.h and NDPluginProcess.h through records in NDPluginBase.template and NDProcess.template. In this example the input image is first offset by -4 and scaled by 35. That result is then clipped to a minimum of 0 and a maximum of 255. The image is then run through a recursive averaging filter with N=100.

NDProcess.adl

The following is the MEDM screen for reading a background image or flatfield image from a TIFF file. It is a stripped-down version of the normal TIFF plugin screen, and allows specifying the path and filename components to construct the full filename.

NDProcessTIFF.adl

Image collected with 30 microsecond exposure time, so it is very noisy. This is the image output from the NDPlugProcess plugin with Offset and Scale as shown above, but with the recursive filter disabled.

NDPluginStats plugin connected to the NDPluginProcess filter for the noisy image above. Note that there are only 6 non-zero intensities in the histogram, because the brightest pixels are only 6 A/D units above background.

Same image collected with 30 microsecond exposure time. This is the image output from the NDPlugProcess plugin with Offset and Scale as shown above, with the recursive filter enabled. Note the dramatic improvement in signal to noise.

NDPluginStats plugin connected to the NDPluginProcess filter for the filtered image above. Note that there are many non-zero intensities in the histogram, because the filtering is improving the dynamic range significantly.