使用直方图进行图像匹配

使用直方图进行图像匹配

---

前言

图像直方图是用来表现图像中亮度分布的直方图，给出的是图像中某个亮度或者某个范围亮度下共有几个像素，即统计一幅图某个亮度像素数量。

---

提示：以下是本篇文章正文内容，下面案例可供参考

一、直方图匹配（使用函数）

使用cv2.calcHist()、cv2.compareHist(）函数
代码如下（示例）：

import cv2
import numpy as np

# 读取图像
whole_image = cv2.imread("/Pictures/hawen.jpg")
part_image = cv2.imread("/Pictures/cut_hawen.jpg")

# 将图像转换为灰度图像
gray_whole_image = cv2.cvtColor(whole_image, cv2.COLOR_BGR2GRAY)
gray_part_image = cv2.cvtColor(part_image, cv2.COLOR_BGR2GRAY)
# print(gray_part_image.shape)

# 计算部分图像的直方图
hist_part = cv2.calcHist([gray_part_image], [0], None, [256], [0, 256])
# print(hist_part)

# 获取部分图像的尺寸
part_height, part_width = gray_part_image.shape[:2]
# print(part_height, part_width)

# 获取全部图像的尺寸
whole_height, whole_width = gray_whole_image.shape[:2]
# print(whole_height, whole_width)

best_np = []
best_similarity = -100
x, y = 0, 0

# 循环切割图片
for i in range(whole_height - part_height):
    for j in range(whole_width - part_width):
        # 切割图片
        cut_image = whole_image[i:i + part_height, j:j + part_width]
        # 转为灰度
        gray_cut_image = cv2.cvtColor(cut_image, cv2.COLOR_BGR2GRAY)
        # 计算切割图像的直方图
        hist_cut = cv2.calcHist([gray_cut_image], [0], None, [256], [0, 256])
        # 使用巴氏距离比较直方图相似度
        similarity = cv2.compareHist(hist_part, hist_cut, cv2.HISTCMP_CORREL)
        if similarity > best_similarity:
            best_similarity = similarity
            best_np = cut_image
            x = j
            y = i

# 画图
cv2.rectangle(whole_image, (x, y), (x + part_width, y + part_height), (0, 255, 0), 1)

# 显示结果
cv2.imshow('Matching Result', whole_image)
cv2.waitKey(0)
cv2.destroyAllWindows()

二、直方图匹配（函数实现版）

自己编写cv2.calcHist()、cv2.compareHist(）函数
代码如下（示例）：

import cv2
import numpy as np
import math


def calcHist(image):
    hist = np.zeros(256)
    hist.shape = (256, 1)
    for data in image:
        # print(data)
        for num in data:
            hist[num] = hist[num] + 1.0

    return hist


def compareHist(h1, h2):
    sum1 = 0
    sum2 = 0
    for data1 in h1:
        # print(data)
        for num1 in data1:
            sum1 = sum1 + num1
    avg1 = sum1 / h1.shape[0]
    # print(avg1)
    for data2 in h2:
        # print(data)
        for num2 in data2:
            sum2 = sum2 + num2
    avg2 = sum2 / h2.shape[0]
    # print(h1[0][0])
    up = 0
    for i in range(h1.shape[0]):
        up = up + (h1[i][0] - avg1) * (h2[i][0] - avg2)

    down = 0
    one = 0
    two = 0
    for i in range(h1.shape[0]):
        one = one + (h1[i][0] - avg1) * (h1[i][0] - avg1)
        two = two + (h2[i][0] - avg2) * (h2[i][0] - avg2)
    re = one * two
    down = math.sqrt(re)
    return up / down


# 读取图像
whole_image = cv2.imread("/Pictures/hawen.jpg")
part_image = cv2.imread("/Pictures/cut_hawen.jpg")

# 将图像转换为灰度图像
gray_whole_image = cv2.cvtColor(whole_image, cv2.COLOR_BGR2GRAY)
gray_part_image = cv2.cvtColor(part_image, cv2.COLOR_BGR2GRAY)

# 计算部分图像的直方图
hist_part = calcHist([gray_part_image])
# print(hist_part.shape)
# print(type(hist_part))

# 获取部分图像的尺寸
part_height, part_width = gray_part_image.shape[:2]
# print(part_height, part_width)

# 获取全部图像的尺寸
whole_height, whole_width = gray_whole_image.shape[:2]
# print(whole_height, whole_width)

best_np = []
best_similarity = -100
x, y = 0, 0

# 循环切割图片
for i in range(whole_height - part_height):
    for j in range(whole_width - part_width):
        print(i,j)
        # 切割图片
        cut_image = whole_image[i:i + part_height, j:j + part_width]
        # 转为灰度
        gray_cut_image = cv2.cvtColor(cut_image, cv2.COLOR_BGR2GRAY)
        # 计算切割图像的直方图

        hist_cut = calcHist([gray_cut_image])
        # print(hist_cut.shape)
        # print(type(hist_cut))

        # 使用巴氏距离比较直方图相似度
        similarity = compareHist(hist_part, hist_cut)
        if similarity > best_similarity:
            best_similarity = similarity
            best_np = cut_image
            x = j
            y = i

# 画图
cv2.rectangle(whole_image, (x, y), (x + part_width, y + part_height), (0, 255, 0), 1)

# 显示结果
cv2.imshow('Matching Result', whole_image)
cv2.waitKey(0)
cv2.destroyAllWindows()

---

总结

以上就是今天要讲的内容，本文仅仅简单介绍了使用直方图进行图像匹配，请各路大佬多多指正！