python 图片融合HSV/IHS/PCA

1 HSV

HSV可进行RGB图像到HSV色度空间的变换，用高分辨率的图像代替颜色亮度值波段，自动用最近邻、双线性或三次卷积技术将色度和饱和度重采样到高分辨率像元尺寸，然后再将图像变换回RGB色度空间。输出的RGB图像的像元将与高分辨率数据的像元大小相同。

# https://blog.csdn.net/weixin_45364562/article/details/107130219
def hsv_imgfusion(low, high):
    """
    HSV融合低分辨率和高分辨率图像 H色调,S饱和度,V明度
    输入：图像三维数组
    返回：可绘出图像的utf-8格式的三维数组
    """
    hgt, wth = high.shape[:2]
    data_low = cv2.resize(low, (wth, hgt), interpolation=cv2.INTER_CUBIC)  # 重采样

    h_low, s_low, v_low = cv2.split(
        cv2.cvtColor(
            data_low, cv2.COLOR_BGR2HSV
        )
    )

    h_high, s_high, v_high = cv2.split(
        cv2.cvtColor(
            high, cv2.COLOR_BGR2HSV
        )
    )

    # 用高分影像的V替换低分影像的V
    HSV = cv2.merge([h_low, s_low, v_high])
    RGB = Image.fromarray(
        cv2.cvtColor(
            HSV, cv2.COLOR_HSV2RGB
        )
    )
    return RGB

结果如下：左边是高分辨率，中间是低分辨率，右边是融合图像。

2 IHS

利用IHS变换进行融合，通过正反变换矩阵操作，用高分辨率图像的第一波段替换低分辨率图像，以达到融合效果。

# https://blog.csdn.net/weixin_45364562/article/details/107130857
def ihs_imgfusion(data_l, data_h):
    """
    基于IHS变换融合算法
    输入：np.ndArray格式的三维数组
    返回：可绘出图像的utf-8格式的三维数组
    """

    # RGB－>IHS正变换矩阵
    A = [[1. / 3., 1. / 3., 1. / 3.],
         [-np.sqrt(2) / 6., -np.sqrt(2) / 6., 2 * np.sqrt(2) / 6],
         [1. / np.sqrt(2), -1. / np.sqrt(2), 0.]]

    # IHS－>RGB逆变换矩阵
    B = [[1., -1. / np.sqrt(2), 1. / np.sqrt(2)],
         [1., -1. / np.sqrt(2), -1. / np.sqrt(2)],
         [1., np.sqrt(2), 0.]]
    A = np.matrix(A)
    B = np.matrix(B)

    # 输入数据参数
    band, w, h = data_h.shape
    data_low = data_l.reshape(3, w * h)
    data_high = data_h.reshape(3, w * h)

    # 正变换
    a1 = np.dot(A, np.matrix(data_high))
    a2 = np.dot(A, np.matrix(data_low))

    # 用高分影像第一波段替换低分影像第一波段
    a2[0, :] = a1[0, :]

    # 融合影像逆变换
    rgb = np.array(np.dot(B, a2))

    rgb = rgb.reshape((3, w, h))

    min_val = np.min(rgb.ravel())
    max_val = np.max(rgb.ravel())

    rgb = np.uint8((rgb.astype(np.float) - min_val) / (max_val - min_val) * 255)
    rgb = Image.fromarray(cv2.merge([rgb[0], rgb[1], rgb[2]]))
    return rgb

3 PCA

融合流程如下：

def PCA_fusion(data_l, data_h):
    imageSize = data_l.size
    # Todo: for more than two images
    allImage = np.concatenate((data_h.reshape(1, imageSize), data_l.reshape(1, imageSize)), axis=0)
    covImage = np.cov(allImage)
    D, V = np.linalg.eig(covImage)
    D = D.tolist()
    LD = max(D)
    if D.index(LD) == 0:
        a = V[:, 0] / V[:, 0].sum()
    elif D.index(LD) == 1:
        a = V[:, 1] / V[:, 1].sum()
    elif D.index(LD) == 2:
        a = V[:, 2] / V[:, 2].sum()
    elif D.index(LD) == 3:
        a = V[:, 3] / V[:, 3].sum()
    fusionImage = data_h[0]*a[0] + data_l*a[1]
    fusionImage = fusionImage.swapaxes(1, 0)
    fusionImage = fusionImage.swapaxes(2, 1)
    fusionImage = np.uint8(fusionImage)
    return fusionImage

参考：
https://blog.csdn.net/iteye_16040/article/details/82028976
https://blog.csdn.net/weixin_44588584/article/details/129798346
https://blog.csdn.net/LiuKY_123/article/details/124135725
https://blog.csdn.net/nanke_yh/article/details/89481095
https://www.cnblogs.com/suoyike1001/p/15248811.html
https://blog.csdn.net/qq_41730491/article/details/128933113