op normalize

Mathematically, normalization scales the intensity of image pixels relative to a given normalization value (typically 1) and type (typically NORM_L2). Normalization provides a valuable standard of comparison and is implemented by OpenCV normalize.

Normalization can also be used for contrast shifting and scaling to focus on "pixel values of interest" and is a great visual aid for interpreting images. FireSight extends OpenCV with a domain parameter that lets you choose a range of pixel intensities of interest (see Example 3).

• alpha Norm value to normalize to or the lower range boundary in case of the range normalization. Default is 1 unless image is 8-bit, in which case FireSight will automatically compute a good value.
• beta Upper range boundary in case of the range normalization; it is not used for the norm normalization. Only used for NORM_MINMAX.
• normType normalization type (NORM_L2 (default), NORM_L1, NORM_MINMAX, NORM_INF).

For 8-bit images, FireSight provides two parameters which are much more convenient to use than alpha and beta:

• domain JSON array of integers describing value interval to normalize. Default is [0,255]. Values outside this interval will be eliminated from pipeline image by truncating and shifting image values. This works for all normalization types and is useful for focusing on a value interval of interest.
• range JSON array of integers describing output value interval. Default is [0,255], which maximizes the dynamic range of the normalized output. This is rarely used.

Stage Model

{}

Examples

The examples below use the following test image. The test image has pixel values in the range [0,255] with a background pixel value of 32.
Within the test image are letters in the grayscale range [20,52]. The letters are difficult to see with the naked eye:

Example 1: Show ABC pipeline

firesight -i img/abc.png -p json/normalize.json -o target/normalize.png

Pixel:0.6ms RPi:11ms

The default NORM_L2 normalization reveals the letters hidden in the test image.

Example 2: High Contrast ABC pipeline

firesight -i img/abc.png -p json/normalize.json -o target/normalize.png -Ddomain=[20,52]

Pixel:0.8ms RPi:14ms

By choosing a domain that matches the values used to represent the letters, we eliminate uninteresting dark/light values. This allows us to increase the image gain and bias it toward the value interval of interest:

See Also

• op absdiff
• op threshold

Example 3: normType and domain

The following table illustrates the utility of the various normType options. It also shows how the domain interval can improve contrast.

normType	no domain	domain=[20,52]	no domain	domain=[164,196]
NORM_MINMAX
NORM_INF
NORM_L1
NORM_L2