Skip to main content
Images are naturally represented as multi-dimensional arrays. A grayscale image is a 2-D array of pixel intensities, while a color image is a 3-D array with a channel dimension. numpy-ts provides the array operations needed for common image processing tasks.

​
Representing images as 2D/3D arrays

​
Slicing regions of interest

Extract sub-images using NumPy-style slicing.

​
Element-wise transforms: brightness and contrast

Adjust pixel values using vectorized arithmetic — no loops required.

​
Reshaping and transposing for channel reordering

Different frameworks expect different axis orderings. Use transpose and reshape to convert between them.

​
Simple convolution filter

Apply spatial filters to images using convolution. For a 2-D image, convolve each row and column separately using 1-D convolution, or build a simple approach using element-wise operations.
For large-scale 2-D convolution, consider using the FFT approach: pad the image and kernel to the same size, multiply their FFTs, and inverse-transform the result. This runs in O(N log N) instead of O(N * K).