动态规划搜索图像最佳拼接线

花花世界的小憩

于 2023-09-27 17:27:29 发布

阅读量334

点赞数 1

文章标签：动态规划 opencv 算法

本文链接：https://blog.csdn.net/weixin_40531064/article/details/133358608

版权

import cv2
import numpy as np
import matplotlib.pyplot as plt

# 加载图像并转换为灰度图像
img1 = cv2.imread("1.jpg")
img2 = cv2.imread("2.jpg")

gray1 = cv2.cvtColor(img1, cv2.COLOR_BGR2GRAY)
gray2 = cv2.cvtColor(img2, cv2.COLOR_BGR2GRAY)

# 高斯滤波
blur1 = cv2.GaussianBlur(gray1, (3, 3), 0)
blur2 = cv2.GaussianBlur(gray2, (3, 3), 0)

# 计算梯度大小矩阵
grad_x1 = cv2.Sobel(blur1, cv2.CV_32F, 1, 0, ksize=3)
grad_y1 = cv2.Sobel(blur1, cv2.CV_32F, 0, 1, ksize=3)
grad_mag1 = cv2.magnitude(grad_x1, grad_y1)
grad_x2 = cv2.Sobel(blur2, cv2.CV_32F, 1, 0, ksize=3)
grad_y2 = cv2.Sobel(blur2, cv2.CV_32F, 0, 1, ksize=3)
grad_mag2 = cv2.magnitude(grad_x2, grad_y2)

# 将梯度大小矩阵转换为能量图像
# energy1 = np.sum(grad_mag1, axis=2)
energy1 = cv2.normalize(grad_mag1, None, alpha=0, beta=255, norm_type=cv2.NORM_MINMAX, dtype=cv2.CV_8U)

# energy2 = np.sum(grad_mag2, axis=2)
energy2 = cv2.normalize(grad_mag2, None, alpha=0, beta=255, norm_type=cv2.NORM_MINMAX, dtype=cv2.CV_8U)

# 动态规划算法选择最佳路径
def find_optimal_seam(energy):
    height, width = energy.shape
    dp = np.zeros_like(energy)
    backtrack = np.zeros_like(energy, dtype=int)

    dp[0, :] = energy[0, :]

    for i in range(1, height):
        for j in range(width):
            # 处理边界情况
            if j == 0:
                indices = [j, j + 1]
            elif j == width - 1:
                indices = [j - 1, j]
            else:
                indices = [j - 1, j, j + 1]

            min_index = min(indices, key=lambda x: dp[i - 1, x])
            dp[i, j] = energy[i, j] + dp[i - 1, min_index]
            backtrack[i, j] = min_index

    # 寻找最佳路径的结束位置
    seam_end = np.argmin(dp[-1, :])

    # 回溯最佳路径
    seam = [(height - 1, seam_end)]
    for i in range(height - 2, -1, -1):
        seam_end = backtrack[i + 1, seam_end]
        seam.append((i, seam_end))
    print(seam)
    return seam  # 反转路径顺序，从上到下

seam1 = find_optimal_seam(energy1)
seam2 = find_optimal_seam(energy2)

def stitch_images(img1, img2, seam):
    height, width, _ = img1.shape
    result_height = height + len(seam)
    result_img = np.zeros((result_height, width, 3), dtype=np.uint8)

    # 复制第一张图像的像素值
    result_img[:height, :, :] = img1

    # 创建用于遮挡右侧像素的掩膜
    mask = np.ones((height, width), dtype=np.uint8)
    for i, j in seam:
        mask[i, j:] = 0

    # 根据缝合线从第二张图像中选择像素值并使用掩膜覆盖右侧像素
    seam_indices = np.array([(i, j) for i, j in seam])
    result_img[seam_indices[:, 0] + height, seam_indices[:, 1], :] = img2[seam_indices[:, 0], seam_indices[:, 1], :]
    mask[seam_indices[:, 0], seam_indices[:, 1]] = 0
    result_img[height:, :, :] *= mask[:, :, np.newaxis]

    return result_img

result_img = stitch_images(img1, img2, seam1)
def energy_to_image(energy):
    # 将能量值转换为颜色
    color_energy = cv2.applyColorMap(energy, cv2.COLORMAP_JET)
    # 转置energy的矩阵，使得能量图像较长的那个边朝向横轴
    color_energy =np.array(color_energy)
    return color_energy


def stitch_images(img1, img2, seam):
    height, width, _ = img1.shape

    # 获取拼接线的最大列数（即最右侧的列数）
    max_seam_col = max([j for _, j in seam])

    # 创建结果矩阵，宽度为两幅图像的宽度之和减去拼接线的最大列数
    result_width = width + (img2.shape[1] - max_seam_col)
    result_img = np.zeros((height, result_width, 3), dtype=np.uint8)

    # 将第一张图像复制到结果矩阵中
    result_img[:, :width, :] = img1

    # 将第二张图像复制到结果矩阵中，仅保留拼接线左侧的部分
    result_img[:, width:, :] = img2[:, max_seam_col :, :]

    return result_img


# 绘制能量图像以及拼接线
energy_img1 = energy_to_image(energy1)
energy_img2 = energy_to_image(energy2)

plt.subplot(2, 1, 1)
plt.imshow(energy_img1)
plt.plot([j for _, j in seam1], [i for i, _ in seam1], color='r', linewidth=2)

plt.subplot(2, 1, 2)
plt.imshow(energy_img2)
plt.plot([j for _, j in seam2], [i for i, _ in seam2], color='r', linewidth=2)
plt.show()
# 拼接两幅图像
result_img = stitch_images(img1, img2, seam1)

# 显示拼接结果
plt.imshow(cv2.cvtColor(result_img, cv2.COLOR_BGR2RGB))
plt.title('Stitched Image')
plt.show()

传统动态规划算法

第一步：假设能量函数图中第一行各列像素点都对应一条缝合线，其能量值大小均初始化为其对应的当前点能量值；

第二步：从第二行开始，为其每一点在上一行中选取一个最佳路径节点，选取方式的具体化方法是比较当前点正对的上一行中3个邻近点的能量值，将其中的最小值对应的列记录下来，并将该最小值与当前点对应的能量值相加来更新缝合线的能量值：