Python 实现直方图均衡

Python 直方图均衡

代码

import numpy as np


def histequalize(X):
    if len(X.shape) == 2:   # 判断是否为单通道
        N = X.shape[0] * X.shape[1]
        hist_X = np.bincount(X.reshape((N, )), minlength=256)
        XtoY = np.zeros((256, ))
        Y = np.zeros(X.shape)
        for i in range(256):
            XtoY[i] = np.sum(hist_X[0:i]) + hist_X[i]
        XtoY = XtoY / N
        XtoY = np.round(XtoY * 255)    # 如果使用PPT上的公式，结果与opencv不一样
        # XtoY = np.round((XtoY - np.min(XtoY)) / (1 - np.min(XtoY)) * 255)  # 对映射XtoY进行归一化，得到的结果与opencv相同
        for i in range(X.shape[0]):
            for j in range(X.shape[1]):
                Y[i, j] = XtoY[X[i, j]]
        Y = Y.astype('int32')
        hist_Y, _ = np.histogram(Y, 255, (0, 255))
        return Y, hist_Y
    else:
        return

测试

对如下矩阵进行均衡化

[[1, 5, 255, 225, 100, 200, 255, 200],
[1, 7, 254, 255, 100, 10, 10, 9],
[3, 7, 10, 100, 100, 2, 9, 6],
[3, 6, 10, 10, 9, 2, 8, 2],
[2, 1, 8, 8, 9, 3, 4, 2],
[1, 0, 7, 8, 8, 3, 2, 1],
[1, 1, 8, 8, 7, 2, 2, 1],
[2, 3, 9, 8, 7, 2, 2, 0]]

• 如果按照一般的公式，结果与opencv库函数的结果有差别。对应代码为

XtoY = np.round(XtoY * 255)

结果为

	[[ 40 112 255 239 227 235 255 235]
	[ 40 139 243 255 227 211 211 191]
	[104 139 211 227 227  84 191 120]
	[104 120 211 211 191  84 171  84]
	[ 84  40 171 171 191 104 108  84]
	[ 40   8 139 171 171 104  84  40]
	[ 40  40 171 171 139  84  84  40]
	[ 84 104 191 171 139  84  84   8]]

• 如果对映射公式进行归一化，结果与opencv库函数相同。对应代码为

 XtoY = np.round((XtoY - np.min(XtoY)) / (1 - np.min(XtoY)) * 255)

结果为

[[ 33 107 255 239 226 234 255 234]
	[ 33 136 243 255 226 210 210 189]
	[ 99 136 210 226 226  78 189 115]
	[ 99 115 210 210 189  78 169  78]
	[ 78  33 169 169 189  99 103  78]
	[ 33   0 136 169 169  99  78  33]
	[ 33  33 169 169 136  78  78  33]
	[ 78  99 189 169 136  78  78   0]]