图像处理之三种经典插值方法

原理介绍参见：
一个
二个

import numpy as np
import cv2
import  math

def nearest(im,dst_size):
    '''
    最近邻插值（适用于灰度图）
    :param im: 输入图像
    :param 输出图像的尺寸(h, w, c)
    :return: 缩放后的图像
    '''
    dst = np.zeros(dst_size)
    dst_h = dst_size[0]
    dst_w = dst_size[1]
    sf_h =  im.shape[0]/dst_size[0]
    sf_w =  im.shape[1]/dst_size[1]
    for i in range(dst_h):
        for j in range(dst_w):
            src_i = int(sf_h*i)
            src_j = int(sf_w*j)
            dst[i,j] = im[src_i,src_j]
    return dst

def double_linear(im,dst_size):
	'''
    双线性插值（适用于灰度图）
    :param im: 输入图像
    :param 输出图像的尺寸(h, w, c)
    :return: 缩放后的图像
    '''
    dst = np.zeros(dst_size)
    dst_h = dst_size[0]
    dst_w = dst_size[1]
    sf_h =  im.shape[0]/dst_size[0]
    sf_w =  im.shape[1]/dst_size[1]
    for i in range(dst_h):
        for j in range(dst_w):
            src_i = sf_h*i
            src_j = sf_w*j
            x1 = math.floor(src_i)
            y1 = math.floor(src_j)
            u = src_i-x1
            v = src_j-y1
            x2 = x1+1
            y2 = y1
            x3 = x1
            y3 = y1+1
            x4 = x1+1
            y4 = y1+1
            if x1+1>=im.shape[0]:
                x4 = im.shape[0] - 1
                x2 = x4
                x1 = x4 - 1
                x3 = x4 - 1
            if y1+1>=im.shape[1]:
                y4 = im.shape[1] - 1
                y3 = y4
                y1 = y4 - 1
                y2 = y4 - 1
            # dst[i,j] = im[src_i,src_j]
            dst[i, j] = (1 - u) * (1 - v) * im[x1, y1] + (1 - u) * v * im[x2, y2] + u * (1 - v) * im[x3, y3] + u * v * im[x4, y4]
    return dst

def Bicubic_Bell( num ):
   # print( num)
    if  -1.5 <= num <= -0.5:
      #  print( -0.5 * ( num + 1.5) ** 2 )
        return -0.5 * ( num + 1.5) ** 2
    if -0.5 < num <= 0.5:
       # print( 3/4 - num ** 2 )
        return 3/4 - num ** 2
    if 0.5 < num <= 1.5:
       # print( 0.5 * ( num - 1.5 ) ** 2 )
        return 0.5 * ( num - 1.5 ) ** 2
    else:
       # print( 0 )
        return 0
def Bicubic(img, bigger_height, bigger_width, channels):
	'''
    最三次插值（适用于灰度图）
    :param img: 输入图像
    :param bigger_height:输出图像的h
    :      bigger_width:输出图像的w
    :      channel:  输出图像的c
    :return: 缩放后的图像
    '''
    Bicubic_img = np.zeros(shape=(bigger_height, bigger_width, channels), dtype=np.uint8)

    for i in range(0, bigger_height):
        for j in range(0, bigger_width):
            row = (i / bigger_height) * img.shape[0]
            col = (j / bigger_width) * img.shape[1]
            row_int = int(row)
            col_int = int(col)
            u = row - row_int
            v = col - col_int
            tmp = 0
            for m in range(-1, 3):
                for n in range(-1, 3):
                    if (row_int + m) < 0 or (col_int + n) < 0 or (row_int + m) >= img.shape[0] or (col_int + n) >= \
                            img.shape[1]:
                        row_int = img.shape[0] - 1 - m
                        col_int = img.shape[1] - 1 - n

                    numm = img[row_int + m][col_int + n] * Bicubic_Bell(m - u) * Bicubic_Bell(n - v)
                    tmp += np.abs(np.trunc(numm))

            Bicubic_img[i][j] = tmp
    return Bicubic_img
if __name__=='__main__':
    im = cv2.imread(r'E:\\1.png')#,flags=cv2.IMREAD_GRAYSCALE)
    cv2.imshow('Image',im)
    cv2.waitKey(0)
    # res = double_linear(im,(300,300,3))
    # res = nearest(im, (300, 300, 3))
    # res = nearest_neighbor(im, 2)
    res = Bicubic(im, 300, 300, 3)
    cv2.imshow('Image', np.uint8(res))
    cv2.waitKey(0)