Procrustes Alignment Error

mediumPROLeast Squares AlignmentAlignmentCV: Image Alignment and Stitching

You are given two sets of corresponding 2D points and need to compute the alignment error (sum of squared distances between corresponding points).

After aligning point sets (e.g., by centering at origin), the alignment error measures how well they match:

$E = \sum_{i=1}^{n} \|p_i - q_i\|^2 = \sum_{i=1}^{n} [(p_{x,i} - q_{x,i})^2 + (p_{y,i} - q_{y,i})^2]$

This is used in:

Procrustes analysis for shape matching
ICP convergence checking
Evaluating transformation quality

Lower error means better alignment.

Example

Input

points1 = [(0, 0), (1, 0)]
points2 = [(0, 0), (1, 0.1)]

Output

0.01

Reasoning

Inputs: points1 and points2 each hold $n = 2$ points in the plane, paired by position: $p_1 = (0, 0)$ goes with $q_1 = (0, 0)$ , and $p_2 = (1, 0)$ goes with $q_2 = (1, 0.1)$ .
Pair 1: the points coincide, so both coordinate differences are $0$ and $\|p_1 - q_1\|^2 = 0^2 + 0^2 = 0$ .
Pair 2: the $x$ difference is $1 - 1 = 0$ and the $y$ difference is $0 - 0.1 = -0.1$ , so $\|p_2 - q_2\|^2 = 0^2 + (-0.1)^2 = 0.01$ . Squaring makes the sign irrelevant.
Total: $E = 0 + 0.01 = 0.01$ , which stays 0.01 after rounding to 4 decimal places (rounding also cleans up tiny floating point noise in $0.1^2$ ).
Notice: the error is a squared distance, not a distance. The second point is off by $0.1$ units, yet contributes only $0.01$ ; an offset of $(1, 1)$ would contribute $1^2 + 1^2 = 2$ . Identical sets give exactly $0$ .

Constraints

points1 and points2 are lists of corresponding (x, y) points of same length
Each point in points1 corresponds to the same-indexed point in points2
Return total squared error rounded to 4 decimal places

More from CV: Image Alignment and Stitching

Jacobian for Bundle Adjustmenthard Cylindrical Projection for Panoramashard Essential Matrix Estimationhard Procrustes Analysismedium Gradient Descent Updatemedium

Concept primerPREVIEW

Procrustes Alignment Error: Background & Strategy

Background Knowledge

Procrustes Analysis Fundamentals

Procrustes analysis is a statistical technique for comparing two shapes by finding the optimal transformation (rotation, reflection, and/or translation) that aligns one point set to another. The core idea is that two shapes may be identical but differ in position, orientation, or scale. By applying geometric transformations, we can measure how similar the shapes truly are. The alignment error quantifies this similarity: it's the sum of squared Euclidean distances between corresponding points after optimal alignment. This metric is fundamental because it's scale-invariant and rotation-invariant once proper alignment is applied.

Why This Matters in Computer Vision

In image alignment and stitching applications, you frequently encounter scenarios where you have two sets of corresponding feature points (e.g., detected corners or keypoints) from different images or views. The Procrustes alignment error helps determine whether these correspondences are correct and how well the geometric transformation between images works. Lower error indicates better alignment quality. This is also critical in Iterative Closest Point (ICP) algorithms, where alignment error convergence signals when the algorithm has found a stable solution.

Mathematical Foundation

The formula you've been given computes the sum of squared residuals after alignment. Before computing this error, the point sets are typically centered (translated so their centroids is at the origin). This preprocessing step removes translation as a variable, allowing you to focus on rotation and scale. The squared distances are summed rather than taking absolute distances because: (1) squaring penalizes larger errors more heavily, (2) it's mathematically convenient for optimization, and (3) it's differentiable everywhere, making it suitable for gradient-based methods.

Algorithm/Approach

The general approach to solving Procrustes alignment error problems follows this pattern:

Preprocess the point sets: Center both sets by subtracting their centroids
Compute the optimal transformation: Use Singular Value Decomposition (SVD) or similar techniques to find the best rotation/reflection matrix
Apply the transformation: Transform one point set using the optimal transformation
Calculate the error: Compute the sum of squared distances using the formula provided

The key insight is that once you've found the optimal alignment transformation, computing the error is straightforward—it's just the Euclidean distance formula applied to all corresponding points.

PRO

Continue the full explanation

You're reading the free preview. Unlock the complete walkthrough, the code editor, test runner and reference solution with Premium.

PRO

Editor locked

The code editor is locked for Pro problems. It is only available for free problems. Please upgrade to gain access to the code editor for all problems.

solution.py

Python 3.13

Test results