Sliding window detection - ML & AI Coding

68. Sliding window detection

medium

General

staff

Build a simple sliding-window object detector that scans an image and returns the best-matching window for a given template. You’ll compute a similarity score at each window location and pick the top result.

The matching score is normalized cross-correlation (NCC):

\text{NCC}(P, T)=\frac{\sum_{i,j}(P_{ij}-\bar{P})(T_{ij}-\bar{T})}{\sqrt{\sum_{i,j}(P_{ij}-\bar{P})^2}\sqrt{\sum_{i,j}(T_{ij}-\bar{T})^2}}

Requirements

Implement the function

python

Rules:

Slide the template over image from top-left to bottom-right using the given stride.
At each position, compute NCC between the current image patch and template.
Return the top-left coordinate of the window with the maximum NCC score.
Do not use any prebuilt template-matching utilities (e.g., OpenCV).
Use only NumPy and built-in Python libraries.

Example

python

Output:

python

Input Signature

Argument	Type
image	np.ndarray
stride	int
template	np.ndarray

Output Signature

Return Name	Type
value	tuple

Constraints

Use NumPy; no OpenCV template-matching functions.
Only fully-contained windows; top-left coordinates returned.
Handle zero-variance denominator without crashing.

Hint 1

Convert image and template to NumPy arrays, and iterate only over valid top-left corners: r in [0..H-h], c in [0..W-w] stepping by stride.

Hint 2

Precompute the template’s mean-centered version T0 = T - T.mean() and its norm ||T0|| once; only the image patch statistics change per window.

Hint 3

For each patch P, compute P0 = P - P.mean(), then NCC = sum(P0*T0) / (||P0||*||T0||). If the denominator is 0 (zero variance patch or template), treat the score as -inf so it can’t win.

Roles

ML Engineer

AI Engineer

Companies

General

Levels

staff

senior

entry

Input Arguments

Edit values below to test with custom inputs

You need tolog in/sign upto run or submit