op threshold
The FireSight wrapper for OpenCV threshold compares each pixel to a threshold value and replaces the pixel with a value determined by the threshold type.
-
type Default value is
THRESH_BINARY. -
thresh Threshold value, normally between 0 and 255. Default is
OTSU, which will add the THRESH_OTSU flag to type for automatic thresholding using Otsu's method. See Example 5. -
maxval For THRESH_BINARY and THRESH_BINARY_INV, the replacement value for the type. Default is
255, which is ideal for creating image masks for 8-bit images. -
gray Default is
true, which converts image to grayscale. Iffalse, each channel will be thresholded, unless Otsu's method is chosen, in which the image is autoatically converted to grayscale. (FireSight only)
FireSight only:
-
gray Default value is
true, which converts input image to grayscale before threshold
{}
Example 1: threshold Pointilism pipeline
firesight -i img/absdiff.png -p json/threshold.json -o target/threshold1-color.png -Dthresh=1 -Dgray=false
Pixel:0.5ms
The input image appears bland and simple →
However, setting a threshold of 1 reveals subtle detail →
Example 2: threshold 64 color pipeline
firesight -i img/absdiff.png -p json/threshold.json -o target/threshold64-color.png -Dthresh=64 -Dgray=false
Pixel:0.5ms
Increasing the threshold to 64 reduces the detail, but accentuates colors →
Example 3: threshold 64 grayscale pipeline
firesight -i img/absdiff.png -p json/threshold.json -o target/threshold64.png -Dthresh=64
Pixel:0.2ms
Converting to grayscale gives the expected mask →
Example 4: threshold type THRESH_BINARY_INV pipeline
firesight -i img/absdiff.png -p json/threshold.json -o target/threshold64-inv.png -Dthresh=64 -Dtype=THRESH_BINARY_INV
Pixel:0.2ms
Changing the threshold type to THRESH_BINARY_INV produces the following →
Example 5: Otsu's Method pipeline
firesight -i img/part1-0.png -p json/threshold.json -o target/threshold-otsu.png -Dthresh=OTSU
Pixel:0.7ms
Otsu's method provides automatic thresholding. As with the preceding examples, the original image is generated via absdiff and has many non-zero background pixel differences:
Using Otsu's method, we can easily calculate a foreground/background threshold.
Otsu's method does take more time (compare Example 4), but may be worth the convenience.
