计算机视觉------重采样与插值

目录

一、下采样

 二、上采样

---

一、下采样

对图像进行下采样，可以得到尺寸较小的图像，最简单直接的就是直接按照一定间隔选取原始图像的像素点，即直接下采样，但是这样做的效果并不好。可以采用高斯预滤波平滑图像，再进行下采样，这样会得到相对比较好的结果。

高斯滤波后的图像

原始图像

直接对原始图像下采样

滤波后下采样

import cv2 as cv
import numpy as np

def simple_downsampling(image, scale_factor):
    height, width = image.shape[:2]
    new_height = int(height * scale_factor)
    new_width = int(width * scale_factor)

    new_image = np.zeros((new_height, new_width, 3), dtype=np.uint8)

    for i in range(new_height):
        for j in range(new_width):
            new_image[i, j] = image[int(i / scale_factor), int(j / scale_factor)]

    return new_image

img = cv.imread('./img.png')
height, width = img.shape[: 2]
print(height, width)
simple_downsampled_img = simple_downsampling(img,0.25)
new_height, new_width = simple_downsampled_img.shape[:2]
cv.imwrite('./simple_downsampling.jpg', simple_downsampled_img)
print(new_height, new_width)

# 高斯滤波
out_path2 = './gs_blurred.jpg'
kernel_size2 = (5, 5)  # 滤波器大小
sigma_x = 0  # X方向的高斯核标准差，若为0，根据kernel_size自动计算
gs_blurred_image = cv.GaussianBlur(img, kernel_size2, sigma_x)
cv.imwrite(out_path2, gs_blurred_image)
gs_downsampling_img = simple_downsampling(gs_blurred_image,0.25)
new_height2, new_width2 = simple_downsampled_img.shape[:2]
cv.imwrite('./gs_downsampling_img.jpg', gs_downsampling_img)
print(new_height2, new_width2)

 二、上采样

对图像上采样并进行插值可得到更大尺寸的图像。插值的方法有很多中，只需修改resize的参数即可以实现不同插值效果。

原始图像

最近邻插值

双线性插值

双三次插值

import cv2 as cv

# 读取图像
img = cv.imread('bee.png')  # 替换成你的图像路径
height, width = img.shape[:2]  # 获取图像的高度和宽度
print(height, width)
# 设置新的宽度和高度
new_width, new_height = 350,350  # 设置新的宽度和高度

# 最近邻插值
resized_image_INTER_NEAREST = cv.resize(img, (new_width, new_height), interpolation=cv.INTER_NEAREST)
cv.imwrite('./resized_image_INTER_NEAREST.png', resized_image_INTER_NEAREST)

# 调整图像大小并使用双线性插值
resized_image_INTER_LINEAR = cv.resize(img, (new_width, new_height), interpolation=cv.INTER_LINEAR)
cv.imwrite('./resized_image_INTER_LINEAR.png', resized_image_INTER_LINEAR)

# 双三次插值
resized_image_INTER_CUBIC = cv.resize(img, (new_width, new_height), interpolation=cv.INTER_CUBIC)
cv.imwrite('./resized_image_INTER_CUBIC.png', resized_image_INTER_CUBIC)