Creating an ASCII‑Art Video from a Celebrity Clip Using Python and OpenCV

This article explains how to turn a short video (e.g., a clip of singer Cyndi Wang) into an ASCII‑art video by leveraging Python libraries such as OpenCV, Pillow (PIL), and NumPy. It starts with a brief nostalgic introduction and presents the overall workflow.

Tools required

Development IDE: PyCharm

Runtime: Python 3.7 on Windows 10

Libraries: Pillow (PIL), OpenCV (cv2), NumPy

Project effect

Project idea

The process consists of three main stages: (1) split the source video into individual frames, (2) convert each frame into an ASCII‑style image, and (3) re‑assemble the processed frames into a new video.

Step 1 – Video to frames

<code># 将视频转换为图片 并进行计数，返回总共生成了多少张图片！
def video_to_pic(vp):
    # vp = cv2.VideoCapture(video_path)
    number = 0
    if vp.isOpened():
        r, frame = vp.read()
        if not os.path.exists('cache_pic'):
            os.mkdir('cache_pic')
        os.chdir('cache_pic')
    else:
        r = False
    while r:
        number += 1
        cv2.imwrite(str(number) + '.jpg', frame)
        r, frame = vp.read()
    print('\n由视频一共生成了{}张图片！'.format(number))
    os.chdir('..')
    return number</code>

Step 2 – Convert frames to ASCII images

<code>def img2strimg(frame, K=3):
    # 读取矩阵的长度 有时返回两个值，有时三个值
    height, width, *_ = frame.shape
    # 颜色空间转化 图片对象， 灰度处理
    frame_gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
    # 转换数据类型，将数据降维
    frame_array = np.float32(frame_gray.reshape(-1))
    # K‑means 聚类
    compactness, labels, centroids = cv2.kmeans(frame_array, K, None,
        (cv2.TERM_CRITERIA_EPS + cv2.TERM_CRITERIA_MAX_ITER, 10, 1.0),
        10, cv2.KMEANS_RANDOM_CENTERS)
    centroids = np.uint8(centroids).flatten()
    centroids_sorted = sorted(centroids)
    centroids_index = np.array([centroids_sorted.index(v) for v in centroids])
    # 亮度设置
    bright = [abs((3 * i - 2 * K) / (3 * K)) for i in range(1, K+1)]
    bright_bound = bright.index(np.min(bright))
    # 背景阴影设置
    shadow = [abs((3 * i - K) / (3 * K)) for i in range(1, K+1)]
    shadow_bound = shadow.index(np.min(shadow))
    labels = labels.flatten()
    labels = centroids_index[labels]
    # 每2*2像素挑选一个
    labels_picked = [labels[rows * width:(rows+1)*width:2] for rows in range(0, height, 2)]
    canvas = np.zeros((3 * height, 3 * width, 3), np.uint8)
    canvas.fill(255)
    y = 0
    for rows in labels_picked:
        x = 0
        for cols in rows:
            if cols <= shadow_bound:
                cv2.putText(canvas, str(random.randint(2,9)), (x, y),
                    cv2.FONT_HERSHEY_PLAIN, 0.45, 0.1)
            elif cols <= bright_bound:
                cv2.putText(canvas, '-', (x, y), cv2.FONT_HERSHEY_PLAIN, 0.4, 0, 1)
            x += 6
        y += 6
    return canvas</code>

Step 3 – Assemble ASCII frames into a video

<code>def jpg_to_video(char_image_path, FPS):
    video_fourcc = cv2.VideoWriter_fourcc(*"MP42")  # 使用 MP42 编码器生成更小的视频文件
    char_img_path_list = [char_image_path + r'/{}.jpg'.format(i) for i in range(1, number+1)]
    char_img_test = Image.open(char_img_path_list[1]).size  # 获取分辨率
    if not os.path.exists('video'):
        os.mkdir('video')
    video_writter = cv2.VideoWriter('video/new_char_video.avi', video_fourcc, FPS, char_img_test)
    sum = len(char_img_path_list)
    count = 0
    for image_path in char_img_path_list:
        img = cv2.imread(image_path)
        video_writter.write(img)
        count += 1
        process_bar(count / sum, start_str='', end_str='100%', total_length=15)
    video_writter.release()
    print('\n=======================')
    print('The video is finished!')
    print('=======================')</code>

Simple source sharing

<code># from platypus import
import os
from PIL import Image, ImageFont, ImageDraw
import cv2
import random
import numpy as np
import threading

# Main workflow
if __name__ == '__main__':
    video_path = '王心凌.mp4'
    vp = cv2.VideoCapture(video_path)
    number = video_to_pic(vp)
    for i in range(1, number):
        fp = r"cache_pic/{}.jpg".format(i)
        img = cv2.imread(fp)
        str_img = img2strimg(img)
        cv2.imwrite("cache_char/{}.jpg".format(i), str_img)
    FPS = vp.get(cv2.CAP_PROP_FPS)
    jpg_to_video('cache_char', FPS)</code>

The article concludes with a reminder that the generated video can become large if a high frame rate is used, and provides links to related Python tutorials.