LBP改进算法DLBP

DLBP是一种改进的LBP算法，它引入了一个方向梯度直方图（Directional Gradient Histogram，DGH）来解决LBP算法的旋转不变性问题。DLBP算法的步骤如下：

1.将图像分成若干个小块，对于每个小块，计算其DGH。

2.对于每个像素，计算其梯度方向，并将其分到最近的8个方向之一。

3.对于每个像素，计算其与周围8个像素的灰度值差，并将其转化为一个8位二进制数。

4.将每个像素的二进制数转化为一个十进制数，作为该像素的LBP值。

5.将每个小块中所有像素的LBP值组成一个直方图，作为该小块的DGH。

6.将所有小块的DGH拼接起来，作为整个图像的DLBP特征向量。

下面是DLBP算法的Python实现代码：

import cv2
import numpy as np

def get_dgh(img, cell_size):
    h, w = img.shape[:2]
    gx = cv2.Sobel(img, cv2.CV_32F, 1, 0)
    gy = cv2.Sobel(img, cv2.CV_32F, 0, 1)
    mag, ang = cv2.cartToPolar(gx, gy)
    ang = np.int32(ang / (np.pi / 4) + 0.5) % 8
    dgh = np.zeros((cell_size, cell_size, 8), dtype=np.float32)
    for i in range(cell_size):
        for j in range(cell_size):
            x1 = int(i * h / cell_size)
            x2 = int((i + 1) * h / cell_size)
            y1 = int(j * w / cell_size)
            y2 = int((j + 1) * w / cell_size)
            for k in range(x1, x2):
                for l in range(y1, y2):
                    dgh[i, j, ang[k, l]] += mag[k, l]
    return dgh.flatten()

def get_dlbp(img, cell_size):
    h, w = img.shape[:2]
    lbp = np.zeros((h, w), dtype=np.uint8)
    for i in range(1, h - 1):
        for j in range(1, w - 1):
            center = img[i, j]
            code = 0
            code |= (img[i - 1, j - 1] > center) << 7
            code |= (img[i - 1, j] > center) << 6
            code |= (img[i - 1, j + 1] > center) << 5
            code |= (img[i, j + 1] > center) << 4
            code |= (img[i + 1, j + 1] > center) << 3
            code |= (img[i + 1, j] > center) << 2
            code |= (img[i + 1, j - 1] > center) << 1
            code |= (img[i, j - 1] > center) << 0
            lbp[i, j] = code
    dlbp = np.zeros((cell_size * cell_size * 8), dtype=np.float32)
    for i in range(cell_size, h - cell_size, cell_size):
        for j in range(cell_size, w - cell_size, cell_size):
            dgh = get_dgh(lbp[i - cell_size:i + cell_size, j - cell_size:j + cell_size], cell_size)
            dlbp[(i // cell_size - 1) * (w // cell_size - 2) * 8 + (j // cell_size - 1) * 8:(i // cell_size - 1) * (w // cell_size - 2) * 8 + (j // cell_size - 1) * 8 + 8] = dgh
    return dlbp

img = cv2.imread('test.jpg', cv2.IMREAD_GRAYSCALE)
dlbp = get_dlbp(img, 8)
print(dlbp)