实现OpenCV compareHist：离散直方图输入格式咨询

Hey there! Let's break down exactly how to prep your discrete histograms for OpenCV's compareHist function—this is a super common gotcha with the function's input requirements.

First, let's cut to the chase: compareHist doesn't care about your bin labels directly. It only needs flat, ordered arrays of frequency values from both histograms, aligned perfectly by their bin indices.

Required Input Structure #

Both histograms need to be:

• 1-dimensional arrays (either NumPy arrays or OpenCV Mat objects)
• Composed solely of the freq values from your table
• Ordered to match their corresponding bin indices (e.g., bin 0's freq first, then bin 1's, bin 2's, etc.—which your existing data already follows)
• Stored as float32 data type (OpenCV's preferred format for histogram operations)

Step-by-Step Conversion Example #

Suppose your first histogram looks like this:

bins freq
0 1000
1 200
2 100
3 400

You'd convert it to a valid input array like this (using NumPy, which plays seamlessly with OpenCV):

import numpy as np
hist1 = np.array([1000, 200, 100, 400], dtype=np.float32)

Repeat this exact process for your second histogram—critical note: both arrays must have the same length (same number of bins). If one histogram has bins the other doesn't, fill those missing bins with a frequency of 0 to align them.

Quick Usage Example #

Once your histograms are formatted correctly, using compareHist is simple. Here's a quick snippet:

import cv2

# Assume hist1 and hist2 are your properly formatted arrays
# Choose a comparison method that fits your use case
similarity_score = cv2.compareHist(hist1, hist2, cv2.HISTCMP_CORREL)
print(f"Histogram similarity: {similarity_score}")

Popular comparison methods include HISTCMP_CHISQR (chi-square distance), HISTCMP_BHATTACHARYYA (Bhattacharyya distance), and HISTCMP_INTERSECT (intersection method).

内容的提问来源于stack exchange，提问作者kncdwn