图像质量评价-2

图像质量评价

• 无参考
• 有参考

无参考图像质量评价

1、Laplacian梯度
$$D(f)={\Sigma }_y{\Sigma }_x|G(x,y)|$$
$G(x,y)$是点$(x,y)$处Laplacian算子的卷积。

Laplacian算子，
$$L=\frac{1}{6}\times \left[\begin{array}{ccc}1& 4& 1\\ 4& -20& 4\\ 1& 4& 1\end{array}\right]$$

# 1、Laplacian梯度
def laplacian(img):
    grad = cv2.Laplacian(img, cv2.CV_64F).var()
    return grad

2、Brenner梯度
$$D(f)={\Sigma }_y{\Sigma }_x|f(x+2,y)-f(x,y){|}^2$$

# 2、Brenner梯度
def brenner(img):
    h, w = img.shape
    grad = 0
    for i in range(w - 2):
        for j in range(h):
            grad += (int(img[i + 2, j]) - int(img[i, j])) ** 2
    return grad

3、Tenengrad梯度
$$D(f)={\Sigma }_y{\Sigma }_x|G(x,y)|$$
$G(x,y)$是点$(x,y)$处Sobel算子水平方向和垂直方向梯度的向量。
$$G(x,y)=\sqrt{G_x^2(x,y)+G_y^2(x,y)}$$
$G_x(x,y)$和$G_y(x,y)$是点$(x,y)$处Sobel算子水平方向和垂直方向梯度。

# 3、Tenengrad梯度
def tenengrad(img):
    h, w = img.shape
    grad = 0
    for i in range(w):
        for j in range(h):
            gradX = cv2.Sobel(img, cv2.CV_64F, 1, 0)
            gradY = cv2.Sobel(img, cv2.CV_64F, 0, 1)
            grad += np.sqrt(gradX ** 2 + gradY ** 2)
    return grad

4、Energy梯度
$$D(f)={\Sigma }_y{\Sigma }_x(|f(x+1,y)-f(x,y){|}^2+|f(x,y+1)-f(x,y){|}^2)$$

# 4、Energy梯度
def energy(img):
    h, w = img.shape
    grad = 0
    for i in range(w):
        for j in range(h):
            gradX = (int(img[i + 1, j]) - int(img[i, j])) ** 2
            gradY = (int(img[i, j + 1]) - int(img[i, j])) ** 2
            grad += (gradX + gradY)
    return grad

5、Roberts梯度
$$D(f)={\Sigma }_y{\Sigma }_x(|f(x+1,y+1)-f(x,y){|}^2+|f(x+1,y)-f(x,y+1){|}^2)$$

# 5、Roberts梯度
def roberts(img):
    h, w = img.shape
    grad = 0
    for i in range(w):
        for j in range(h):
            gradX = (int(img[i + 1, j + 1]) - int(img[i, j])) ** 2
            gradY = (int(img[i + 1, j]) - int(img[i, j + 1])) ** 2
            grad += (gradX + gradY)
    return grad

6、SMD
$$D(f)={\Sigma }_y{\Sigma }_x(|f(x,y)-f(x,y-1)|+|f(x,y)-f(x+1,y)|)$$

# 6、SMD函数
def SMD(img):
    h, w = img.shape
    grad = 0
    for i in range(w - 1):
        for j in range(1, h):
            gradX = abs(int(img[i, j]) - int(img[i + 1, j]))
            gradY = abs(int(img[i, j]) - int(img[i, j - 1]))
            grad += (gradX + gradY)
    return grad

7、SMD2
$$D(f)={\Sigma }_y{\Sigma }_x(|f(x,y)-f(x+1,y)|\ast |f(x,y)-f(x,y+1)|)$$

# 7、SMD2函数
def SMD2(img):
    h, w = img.shape
    grad = 0
    for i in range(w - 1):
        for j in range(h - 1):
            gradX = abs(int(img[i, j]) - int(img[i + 1, j]))
            gradY = abs(int(img[i, j]) - int(img[i, j + 1]))
            grad += (gradX * gradY)
    return grad

8、方差
$$D(f)={\Sigma }_y{\Sigma }_x|f(x,y)-\mu {|}^2$$
$\mu$是整个图像像素值的均值。

# 8、方差函数
def variance(img):
    h, w = img.shape
    m = np.mean(img)
    grad = 0
    for i in range(w):
        for j in range(h):
            grad += (img[i, j] - m) ** 2
    return grad

# 这种方法也可以，速度更快
np.var(img)

9、Vollath自相关
$$D(f)={\Sigma }_y{\Sigma }_x(f(x,y)\ast f(x+1,y)-H\ast W\ast {\mu }^2)$$
$H$、$W$分别是图像的高和宽。

# 9、Vollath自相关函数
def vollath(img):
    h, w = img.shape
    m = np.mean(img)
    gradMean = (w + 1) * h * m ** 2
    grad = 0
    for i in range(w - 1):
        for j in range(h):
            grad += img[i, j] * img[i + 1, j]
    return grad - gradMean

10、熵
$$D(f)=-\sum _{i=0}^{L-1}{p}_i\ast lo{g}_2(p_i)$$
$L=256$，表示图像的像素值，$p_i$是图像像素值$i$出现的概率。

# 10、熵函数
def entropy(img):
    h, w = img.shape
    grad = 0
    p = np.bincount(np.array(img).flatten())  # img中每个像素值出现的次数
    for i in range(len(p)):
        if p[i] > 0:
            pi = (p[i] / (w * h))
            grad -= pi * np.log2(pi)
    return grad