使用NIQE对图像进行评价的python代码

NIQE（Natural Image Quality Evaluator）是一种用于自然图像质量评估的算法。下面是使用Python实现NIQE算法的代码：

import numpy as np
import scipy.io as sio
from scipy import ndimage
from scipy.signal import convolve2d

def niqe(img):
    # Load model parameters
    model = sio.loadmat('niqe_model.mat')
    mu_prisparam = model['mu_prisparam']
    cov_prisparam = model['cov_prisparam']
    coef = model['coef']
    q = model['q']
    mu_prisparam = np.squeeze(mu_prisparam)
    cov_prisparam = np.squeeze(cov_prisparam)
    coef = np.squeeze(coef)
    q = np.squeeze(q)

    def im2col_sliding_strided(A, BSZ, stepsize=1):
        m, n = A.shape
        s0, s1 = A.strides
        nrows = m - BSZ[0] + 1
        ncols = n - BSZ[1] + 1
        shp = BSZ[0], BSZ[1], nrows, ncols
        strd = s0, s1, s0, s1

        out_view = np.lib.stride_tricks.as_strided(A, shape=shp, strides=strd)
        return out_view.reshape(BSZ[0] * BSZ[1], -1)[:, ::stepsize]

    # Convert image to grayscale
    if len(img.shape) == 3:
        img = np.dot(img[..., :3], [0.2989, 0.5870, 0.1140])

    # Normalize image
    img = (img - np.mean(img)) / np.std(img)

    # Compute local contrast normalization
    window = np.ones((7, 7)) / 49
    mu = convolve2d(img, window, mode='same', boundary='symm')
    mu_sq = mu * mu
    sigma = np.sqrt(np.abs(convolve2d(img * img, window, mode='same', boundary='symm') - mu_sq))
    structdis = (img - mu) / (1 + sigma)

    # Compute block-wise DCT coefficients
    dct_blocks = []
    for i in range(0, img.shape[0] - 7 + 1, 4):
        for j in range(0, img.shape[1] - 7 + 1, 4):
            block = img[i:i+8, j:j+8]
            dct_block = ndimage.fourier.fctn(block, norm='ortho')
            dct_blocks.append(dct_block)

    # Compute features
    features = []
    for i in range(len(dct_blocks)):
        dct_block = dct_blocks[i]
        mu_dct = np.mean(dct_block)
        sigma_dct = np.std(dct_block)
        features.append(mu_dct)
        features.append(sigma_dct)
        features.append(np.abs(dct_block[0, 1]))
        features.append(np.abs(dct_block[1, 0]))
        features.append(np.mean(np.abs(dct_block[2:, 2:])))
        features.append(np.mean(np.abs(dct_block[3:, 3:])))
        features.append(np.mean(np.abs(dct_block[4:, 4:])))
        features.append(np.mean(np.abs(dct_block[5:, 5:])))

    # Compute feature vector
    features = np.array(features)
    features_norm = (features - mu_prisparam) / np.sqrt(cov_prisparam)

    # Compute quality score
    quality_score = np.dot(coef, np.tanh(np.dot(q, features_norm)))
    return quality_score

使用方法：

import cv2

# Load image
img = cv2.imread('image.jpg')

# Compute NIQE score
niqe_score = niqe(img)
print('NIQE score:', niqe_score)

注意，该代码需要依赖`numpy`、`scipy`和`opencv-python`库。在运行代码前需要先下载NIQE模型文件`niqe_model.mat`，并将其与代码放在同一目录下。