Thresholding in Image Processing

Thresholding in image processing is a segmentation technique used to separate objects from the background by comparing each pixel’s intensity to a threshold value.

If a pixel value is above or below that threshold, it is assigned to a class.

For a simple binary threshold:

$$g(x,y)=\{\begin{array}{ll}1,& f(x,y)\ge T\\ 0,& f(x,y)<T\end{array}$$

Where:

• $f(x,y)$ = original pixel value
• $g(x,y)$ = output pixel value
• $T$ = threshold

Example:

• grayscale pixel = 180, threshold = 128 → output = white
• grayscale pixel = 70, threshold = 128 → output = black

Common types:

• Global thresholding: one threshold for the whole image
• Adaptive thresholding: threshold changes by local region
• Otsu thresholding: automatically finds an optimal threshold

It is commonly used for:

• object detection
• document binarization
• medical image segmentation
• shape extraction