Image Enhancement
Gain-Offset Correction
This determines the contrast in an image. We may have to re-scale images in a video to ensure all images in a sequence have similar brightness and contrast.
This correction fixes two issues of image sensors:
- Gain - A multiplicative error in the output.
- Offset - An additive error in the output.
Scaling & Equalisation
Scaling and equalisation clips and distorts the raw image data to produce a more appealing image. This can reduce the information in the final image. The following methods can be used:
- Histogram Equalisation - Aims to improve apparent contrast.
- Rescales the intensities so that the intensity distribution is near uniform.
Intensity Transforms
Given an image histogram:
- $r$ is the original grey level (between 0 and $L-1$).
- $s$ is the transformed output grey level (also between 0 and $L-1$).
Contrast Stretching
This method is for enhancing images with little contrast.
Given that the detail is given between $r_a$ and $r_b$:
- Reduce the dynamic range before and after the range $r_a$ to $r_b$.
- Increase the contrast in the range $r_a$ to $r_b$ to fill the histogram.
Intensity Level Slicing
This method can be used to enhance features within an image.
Given that the detail is given between $r_a$ and $r_b$:
- Set all pixels in the range to a fixed (brighter) value.
Bit-Plane Slicing
Bit plane slicing is used to separate out useful information within an image:
- In an $n$-bit image, each pixel has $n$ bits and there are $n$ planes.
- The MSB contains the majority of the visually significant data.
- The LSB is essentially noise.
Histogram Stretching
For an ideal image, the image histogram is flat. We can perform histogram stretching to spread a peak to fill the histogram:
- Assume that a histogram of a low contrast image ranges from grey value $\check g$ to $\hat g$.
- Assume that we wish to spread this to $[0,G-1]$.
- where $G-1>\hat g-\check g$.
We can do this transform with the following function:
\[g'=\left\lfloor\frac{g-\check g}{\hat g-\check g}G+\frac12\right\rfloor\]This simple method only stretches the whole histogram. The process of flattening an image histogram is called histogram equalisation.