评价图像模糊度的经典无参考指标的matlab代码

图像模糊度的经典无参考指标包括模糊程度估计（Blind/No-reference Image Blur Assessment，BRISQUE）、清晰度评分（Image Clarity Assessment，ICA）等。以下是 BRISQUE 的 MATLAB 代码实现：

function brisque_score = brisque(image)
    % Load the model parameters
    load('brisqueparameters.mat');
    
    % Convert the input image to grayscale
    if size(image, 3) == 3
        image = rgb2gray(image);
    end
    
    % Compute the feature vector
    features = compute_features(image, mu_prisparam, cov_prisparam, mu_ggdparam, cov_ggdparam);
    
    % Compute the quality score
    brisque_score = features * svm_weights' + svm_bias;
end

function features = compute_features(image, mu_prisparam, cov_prisparam, mu_ggdparam, cov_ggdparam)
    % Compute the image statistics
    mu = mean2(image);
    sigma = std2(image);
    skewness = skewness_func(image(:));
    kurtosis = kurtosis_func(image(:));
    ggd_params = ggd_fit(image(:));
    eta = ggd_params(1);
    beta = ggd_params(2);
    
    % Compute the feature vector
    features = [mu, sigma, skewness, kurtosis, eta, beta];
    for i = 1:length(mu_prisparam)
        features = [features, log(1 + (mu_prisparam(i) - mu)^2 / (cov_prisparam(i) + eps))];
    end
    for i = 1:length(mu_ggdparam)
        features = [features, log(1 + (mu_ggdparam(i) - eta)^2 / (cov_ggdparam(i) + eps))];
    end
end

function skew = skewness_func(x)
    x = double(x);
    n = length(x);
    mu = sum(x) / n;
    sigma = sqrt(sum((x - mu).^2) / n);
    skew = (sum((x - mu).^3) / n) / sigma^3;
end

function kurt = kurtosis_func(x)
    x = double(x);
    n = length(x);
    mu = sum(x) / n;
    sigma = sqrt(sum((x - mu).^2) / n);
    kurt = (sum((x - mu).^4) / n) / sigma^4 - 3;
end

function params = ggd_fit(x)
    x = double(x);
    n = length(x);
    x2 = x.^2;
    sigma2 = mean(x2);
    sigma = sqrt(sigma2);
    gamma_range = 0.2:0.001:10;
    rhat_num = mean(abs(x).^(1/3));
    rhat_den = sqrt(sigma2) * mean(abs(x).^(-1/3));
    rhat = rhat_num / rhat_den;
    [~, index] = min(abs(gamma_range - rhat));
    gamma_hat = gamma_range(index);
    beta_hat = sigma2 / gamma_hat;
    alpha_hat = sqrt(pi) * gamma_hat / (2 * gamma_func(3 / (2 * gamma_hat)));
    params = [alpha_hat, beta_hat];
end

function y = gamma_func(x)
    y = gamma(x);
end

以上代码实现了 BRISQUE 指标的计算，其中包括了特征提取、SVM 模型的权重和偏差、以及 GGD 分布的参数估计等。BRISQUE 的评价范围为 0-100，数值越高则表示图像越清晰。