depth_from_focus — Extract depth using multiple focus levels.
depth_from_focus(MultiFocusImage : Depth, Confidence : Filter, Selection : )
The operator depth_from_focus extracts the depth using a
focus sequence. The images of the focus sequence have to be passed as
a multi channel image (MultiFocusImage). The depth for
each pixel will be returned in Depth as the channel
number. The parameter Confidence returns a confidence
value for each depth estimation: The larger this value, the higher
the confidence of the depth estimation is.
depth_from_focus selects the pixels with the best focus of
all focus levels. The method used to extract these pixels is
specified by the parameters Filter and Selection.
For the parameter Filter, you can choose between the values
'highpass' and 'bandpass'. To determine the focus
within the image a high- or a bandpass filter can be applied. The
larger the filter response, the more in focus is the image at this
location. Compared to the highpass filter, the bandpass filter
suppresses high frequencies. This is useful in particular in images
containing strong noise.
Optionally, you can smooth the filtered image using the mean filter
by passing two additional integer values for the mask size in the
parameter Filter (e.g., ['highpass', 7,
7]). This blurs the in-focus region with neighboring pixels
and thus allows to bridge small areas with no texture within the
image. Note, however, that this smoothing does not suppress noise in
the original image, since it is applied only after high- or bandpass
filtering.
The parameter Selection determines how the optimum focus
level is selected. If you pass the value 'next_maximum',
the closest focus maximum in the neighborhood is used. In contrast,
if you pass the value 'local', the focus level is
determined based on the focus values of all focus levels of the
pixel. With 'next_maximum', you typically achieve a
slightly smoothed and more robust result.
This additional smoothing is useful if no telecentric lenses are used to take the input images. In this case, the position of an object will slightly shift within the sequence. By adding appropriate smoothing, this effect can be partially compensated.
If MultiFocusImage contains more than 255 channels (focus
levels), Depth is clipped at 255, i.e. depth values higher
than 255 are ignored.
If the filter mask for Filter is specified with even values, the
routine uses the next larger odd values instead (this way the center of the
filter mask is always explicitly determined).
If Selection is set to 'local' and Filter is
set to 'highpass' or 'bandpass', depth_from_focus
can be executed on OpenCL devices. If smoothing is enabled, the same
restrictions and limitations as for mean_image apply.
Note that filter operators may return unexpected results if an image with a reduced domain is used as input. Please refer to the chapter Filters.
MultiFocusImage (input_object) multichannel-image(-array) → object (byte)
Multichannel gray image consisting of multiple focus levels.
Depth (output_object) singlechannelimage(-array) → object (byte)
Depth image.
Confidence (output_object) singlechannelimage(-array) → object (byte)
Confidence of depth estimation.
Filter (input_control) string(-array) → (string / integer)
Filter used to find sharp pixels.
Default: 'highpass'
Suggested values: 'highpass', 'bandpass', 3, 5, 7, 9
Selection (input_control) string(-array) → (string)
Method used to find sharp pixels.
Default: 'next_maximum'
List of values: 'local', 'next_maximum'
List of values (for compute devices): 'local'
compose3(Focus0,Focus1,Focus2,&MultiFocus); depth_from_focus(MultiFocus,&Depth,&Confidence,'highpass','next_maximum'); mean_image(Depth,&Smooth,15,15); select_grayvalues_from_channels(MultiChannel,Smooth,SharpImage); threshold(Confidence,HighConfidence,10,255); reduce_domain(SharpImage,HighConfidence,ConfidentSharp);
compose2,
compose3,
compose4,
add_channels,
read_image,
read_sequence
select_grayvalues_from_channels,
mean_image,
binomial_filter,
gauss_filter,
threshold
3D Metrology