Flatten an image into a feature vector

Background

The simplest (if crude) way to feed an image to a dense MLP is to flatten it into a 1-D vector. A 28×28 grayscale image becomes 784 features; an (H, W, C) image becomes $H\cdot W\cdot C$ features, read in row-major order.

Problem statement

Implement flatten_image(img) returning the image as a 1-D np.ndarray.

Input

• img — array-like of shape (H, W) or (H, W, C).

Output

Returns a 1-D np.ndarray of length equal to the number of values in img.

Examples

Example 1 — grayscale

Input:  img = [[1, 2], [3, 4]]
Output: [1 2 3 4]

Example 2 — RGB pixel order (row-major)

Input:  img = [[[1, 2, 3], [4, 5, 6]]]
Output: [1 2 3 4 5 6]

Constraints

• Flatten in row-major (C) order — the NumPy default.
• Return a 1-D array.

Notes

• This is the input a plain MLP would see. It works, but it throws away the 2-D layout, so the model has no built-in notion that neighbouring pixels are related — a key motivation for CNNs.