Concept
DigImage: Particle Tracking
:::: IMAGE CAPTURE
:::: PARTICLE LOCATION
:::: PARTICLE MATCHING
:::: OPERATION
Subsequent Analysis
:::: STRUCTURE OF FILES
:::: BASIC OPERATION OF TRK2DVEL
:::: CONFIGURATION OF TRK2DVEL
:::: FACILITIES
Advanced Features
:::: PARTICLE-TAGGED DATA
:::: IMAGE PREPROCESSING
Experimental Hints
Index
A wide variety of techniques exist for locating features within an image. For DigImage the philosophy has been to keep it simple, partly as sophisticated location strategies do not produce significantly superior results, and partly on the grounds of computational efficiency. It is generally better to start with high quality images of an experiment, than to try to extract data from low quality images by throwing away some of the information.
A popular method of determining where a particle is to fit a Gaussian intensity distribution to the particle image and then determine the centroid of the Gaussian. The philosophical draw back of this approach is that real particles are unlikely to produce Gaussian intensity profiles, particularly as they are probably illuminated from a direction in the plane normal to the viewing direction and hence are illuminated only from one side. Moreover, the image of a particle will generally be relatively small, only a few pixels in size. The amount of information present does not therefore warrant the fitting of a Gaussian profile, especially as some of the information will need to be thrown away.
Any particle location routine should use as much of the available information as possible. Provided the same information and method is used at each successive time for a given particle, the change in the particle position will be represented as accurately as possible, even if the absolute particle position is not quite exact. DigImage employs two methods of locating particles, one which uses all the information, and the other which summarises some of the information. The location of a particle is determined using both these methods, the results being compared to ensure consistency.
A particle in DigImage is defined as an area of an enhanced image (see previous section) satisfying a number of criteria, based on the intensity, size and shape of the particles. The most basic criterion is the intensity which is used to identify potential particles or blobs within an image. For the purposes of this discussion, we shall assume that we are tracking light particles on a dark background. For dark particles on a light background, DigImage inverts the incoming video signal to ensure this is always true.
Blobs are first located by searching for regions within the tracking window of an enhanced image which satisfy a spatially uniform intensity threshold (specified by [;USLU Upper threshold] or determined automatically - see below). As the "background" illumination may have been subtracted, this operation is formally equivalent to searching for regions which have an intensity greater than or equal to the background value plus the threshold. When a region satisfying the threshold criterion has been found, it is marked (as found) and its properties determined to see if it is in fact a particle. DigImage makes a number of passes when searching for blobs. The number of passes is specified by [;USLV Number of threshold levels], each with a lower threshold until the lower threshold specified by [;USLT Lower threshold] is reached. Some classes of blobs are admissible as particles only at the higher threshold levels (e.g. large particles), and some only at the lower threshold levels (e.g. small particles).
As an alternative to specifying the upper threshold, lower threshold and the number of steps, [;USLW Threshold type] allows DigImage to determine the threshold automatically from the intensity histogram of each image as it is processed. In most circumstances better results will be obtained by manual specification of the threshold criteria. However, inexperienced users may find the automatic option easier. The automatic option may also be useful when tracking images whose overall intensity varies with time.
The properties which define a particle are based on its area, shape and intensity distribution. Only blobs which fall within a given range of sizes (based on the pixel area - [;USLR Lower size limit] and [;USLS Upper size limit] - and linear dimensions - [;USLX Minimum horizontal size] and [;USLY Minimum vertical size]), shapes (based on the xy correlation of pixels satisfying the threshold [;USLE Ellipticity limit]) and which have an average intensity sufficiently above the threshold ([;USLA Minimum average intensity excess]) will be treated as particles. Within this range there are a number of categories which indicate how confident (or otherwise) we are that this is a (single or multiple) particle. These categories will be elaborated on in the next section.
The location of the particle is determined from the either the centroid of the area satisfying the prescribed threshold, or by the centroid of the volume (area times intensity) satisfying the same threshold. The difference between these two centroids must be less than some predefined limit ([;USLM Upper limit on centroid mismatch]) before the blob will be treated as a particle. Which measure of the location will be used in the tracking process is up to the user ([;USLC Area/Volume centroid]). The volume centroid is to be preferred as it yields a more accurate position, particularly for particles a few pixels in linear dimension. The accuracy with which the position of the particle is evaluated affects the velocity resolution of the tracking technique. The positional accuracy is a function of the particle size. If the particle is smaller than one pixel, then its position can not be given more accurately than pixel resolution, regardless of which location method is employed. However, under ideal circumstances with a noise-free image and linear response camera, a particle just over one pixel in linear size could be located to an accuracy of 1/256 pixels in that direction. In reality this degree of accuracy is unlikely to be achieved. Nevertheless particles with linear dimensions greater than one pixel, an accuracy of much better than the pixel size may be achieved. Particles should be chosen so that they are as large as possible (in terms of pixels), bearing in mind the assumption that they exactly follow fluid elements (or whatever behaviour is required). One technique for increasing the apparent size of small particles is to defocus the video camera to smear out the particle image. This will be effective, however, only if a very powerful light source were available.
Note also that the greater the use of the available intensity range, the more accurate the volume centroid becomes, provided the image is not saturated.
The options [;USLZ
Count particles in buffer]
and [;USL0 Suggest thresholds]
are provided to help the user determine
appropriate values of the parameters controlling the location
of particles. Typically the specification process will start by
acquiring a few images of the flow from video tape (not directly
from the camera as the intensities tend to be slightly different
from those on the video tape). A first guess at the
thresholds might be made either by looking at the intensity structure
of the image using the cursor (you can use <f5>
to start the cursor menu), from experience and a knowledge of
the colour scheme in use, or by selecting [;USL0
Suggest thresholds] to
provide some idea of appropriate values. The threshold and other
controlling parameters may then be modified by an iterative process
using [;USLZ Count particles in
buffer] to determine which
particles (and how many) are found. A typical strategy will be
to maximise the number of particles found, but simultaneously
minimise the number of blobs rejected.
DigImage documentation page
DigImage home page
Stuart Dalziel's home page