compareHist(HISTCMP_CORREL) returns -1.0 when comparing a histogram with itself

[CloakedLeader]

System Information

• OpenCV: 5.0.0.93 (opencv-python)
• Python: 3.13
• Operating System: Windows

Description

When using cv2.compareHist() with the cv2.HISTCMP_CORREL comparison method, comparing a histogram against itself returns -1.0.

A histogram should have a Pearson correlation coefficient of 1.0 with itself.

This behaviour occurs regardless of whether the histogram is normalized.

Minimal Reproducible Example

import cv2
import numpy as np

print(cv2.__version__)

h = np.random.rand(8, 8, 8).astype(np.float32)

print("Raw:", cv2.compareHist(h, h, cv2.HISTCMP_CORREL))

h1 = cv2.normalize(h, None, alpha=1.0, norm_type=cv2.NORM_L1)
print("L1:", cv2.compareHist(h1, h1, cv2.HISTCMP_CORREL))

h2 = cv2.normalize(h, None, alpha=1.0, norm_type=cv2.NORM_L2)
print("L2:", cv2.compareHist(h2, h2, cv2.HISTCMP_CORREL))

h3 = cv2.normalize(h, None, alpha=0, beta=1, norm_type=cv2.NORM_MINMAX)
print("MINMAX:", cv2.compareHist(h3, h3, cv2.HISTCMP_CORREL))

Actual Output

5.0.0
Raw: -1.0
L1: -1.0
L2: -1.0
MINMAX: -1.0

Expected Output

5.0.0
Raw: 1.0
L1: 1.0
L2: 1.0
MINMAX: 1.0

I also observed this while comparing real image histograms. In that case HISTCMP_CORREL sometimes returned values outside the expected range [-1, 1] (e.g. -2.8) or NaN, while the other comparison methods (HISTCMP_BHATTACHARYYA, HISTCMP_CHISQR, etc.) behaved as expected.