python 图像的特效处理实现

使用python变成实现了六种图像的特殊效果，为：
浮雕、底片、哈哈镜、油画、素描、泛黄
各程序的运行结果均用Lena图像来展示。

一.浮雕特效

大致过程：
使用opencv库的EMBOSS滤波，提取轮廓。
自编写的实现：使用EMBOSS滤波器对应的算子实现。

# 图像特效处理
# 浮雕处理
# 使用opencv函数与自编写函数作效果对比

from PIL import Image
from PIL import ImageFilter
import cv2
import argparse
import numpy as np
from scipy.signal import find_peaks


def image_filters_test():

    im = cv2.imread("Lena.jpg")
    # 预定义的图像增强滤波器
    im_blur = im.filter(ImageFilter.BLUR)  # 模糊滤波
    im_contour = im.filter(ImageFilter.CONTOUR)  # 轮廓滤波
    im_emboss = im.filter(ImageFilter.EMBOSS)  # 浮雕滤波
    im_min = im.filter(ImageFilter.MinFilter(3))  # 最小值滤波器
    im.show()
    im_blur.show()
    im_contour.show()
    im_emboss.show()
    im_min.show()


if __name__ == "__main__":
    # image_filters_test()

    # read in the image and convert to HSV to extract value channel
    img = cv2.imread("Lena.jpg")
    horiz = cv2.flip(img, 1)  # pressing through the pattern flips the image horizontally
    large = cv2.resize(horiz, (0, 0), fx=2, fy=2, interpolation=cv2.INTER_LANCZOS4)
    hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)
    h, s, v = cv2.split(hsv)
    blur = cv2.GaussianBlur(s, (3, 3), 0)
    edge = cv2.Canny(blur, 0, 50)  # perform edge detection with a low slope threshold to capture all edges
    kernel = np.ones((3, 3), np.uint8)
    mask = cv2.dilate(edge, kernel, iterations=1)
    filt = cv2.bitwise_and(cv2.bitwise_not(mask), s)  # filter the noisy edges out by masking off those regions

    # get counts of times each value occurs in the filtered parts of the image
    G = 3
    N = 256 // G + 1
    val_count = [0] * N
    rows, cols = filt.shape
    for i in range(rows):
        for j in range(cols):
            val_count[filt[i, j] // G] += 1
    val_count[0] = 0

    # detect peaks in the histogram, indicating discrete layers
    val_log = np.array([0 if not vc else np.log(vc) for vc in val_count])
    val_norm = [0] * N
    k = 2 * G
    for i in range(N):  # use a windowed z-score to find prominent local maxima
        lo = max(0, i - k)
        hi = min(N, i + k)
        window = val_log[lo:hi]
        val_norm[i] = (val_log[i] - np.mean(window)) / np.std(window)
    val_norm = np.array(val_norm)
    colors, _ = find_peaks(val_norm, height=1, distance=15 // G)

    # separate into layers by color and apply an edge gradient
    # black -> clear to imitate shadows in regions of neg vertical slope
    # white -> clear to imitate highlights in regions of pos vert slope
    result = np.full_like(v, 128)
    upper = np.zeros_like(v)
    for i in range(len(colors) - 1, 0, -1):
        lowc = np.array([0, G * colors[i] - 15, 0])
        highc = np.array([255, G * colors[i]