Face_mosaic - Mosaic blur processing is applied to multiple faces appearing in the video

Last update: Feb 03, 2022

Overview

動機

face_recognitionを使用して得られる顔座標は長方形であり、この座標をそのまま用いてぼかし処理を行った場合得られる画像は醜い。

それに対してモザイクによってぼかし処理を行うと自然に見える。

これを静止画だけでなく動画にも適用したい。
同じ働きを持つものはORB-HD/defaceが既に存在する。defaceがcenterface.onnxを用いるのに比べこちらはdlibを使用する。これらは顔検出のみか顔認識まで行うかで使い分けすると良いと思う。

方法

静止画の場合

face_recognitionのメソッドであるface_recognition.face_locationsの返り値は(top, right, bottom, left)の配列。

顔周りの長方形をぼかす

original.png

import cv2
import face_recognition

image = cv2.imread('img/original.png')
face_location_list = face_recognition.face_locations(
    image, 0, 'cnn')
if len(face_location_list) > 0:
    dst = image.copy()
    for (top, right, bottom, left) in face_location_list:
        dst[top:top + (bottom - top), left:left + (right - left)] = cv2.blur(
            dst[top:top + (bottom - top), left:left + (right - left)], (50, 50))
    cv2.imwrite('img/rectangle_blur.png', dst)

モザイク処理

モザイク処理の方法はnote.nkmk.me Python, OpenCVで画像にモザイク処理（全面、一部、顔など）様よりアイデアを頂戴致しました。

def mosaic(src, ratio):
    small = cv2.resize(src, None, fx=ratio, fy=ratio,
                       interpolation=cv2.INTER_NEAREST)
    return cv2.resize(small, src.shape[:2][::-1], interpolation=cv2.INTER_NEAREST)


if len(face_location_list) > 0:
    dst = image.copy()
    for (top, right, bottom, left) in face_location_list:
        dst[top:top + (bottom - top), left:left + (right - left)] = mosaic(
            dst[top:top + (bottom - top), left:left + (right - left)], 0.1)
    cv2.imwrite('img/rectangle_mosaic.png', dst)

動画に適用

Pixabayが提供するOybek Ostanovの動画をvideo.mp4としてblur処理を行う。

import cv2
import face_recognition
import PySimpleGUI as sg

vcap = cv2.VideoCapture('img/video.mp4', cv2.CAP_FFMPEG)


def mosaic(src, ratio):
    small = cv2.resize(src, None, fx=ratio, fy=ratio,
                       interpolation=cv2.INTER_NEAREST)
    return cv2.resize(small, src.shape[:2][::-1], interpolation=cv2.INTER_NEAREST)


sg.theme('Reddit')
layout = [
    [sg.Text('blur')],
    [sg.Image(key='display')],
    [sg.Button('terminate', key='terminate', button_color='red')]
]
window = sg.Window('blur', layout, location=(50, 50))

while True:
    ret, frame = vcap.read()
    if ret == False:
        break
    face_location_list = face_recognition.face_locations(
        frame, 0, 'cnn')

    if len(face_location_list) > 0:
        dst = frame.copy()
        for (top, right, bottom, left) in face_location_list:
            # rectangle blur ======
            # dst[top:top + (bottom - top), left:left + (right - left)] = cv2.blur(
            #     dst[top:top + (bottom - top), left:left + (right - left)], (50, 50))
            # mosaic blur =========
            dst[top:top + (bottom - top), left:left + (right - left)] = mosaic(
                dst[top:top + (bottom - top), left:left + (right - left)], 0.1)

    event, _ = window.read(timeout=1)
    imgbytes = cv2.imencode(".png", dst)[1].tobytes()
    window["display"].update(data=imgbytes)
    if event == 'terminate':
        break
vcap.release()

結果

期待通りの結果が得られました。

Reference

Author

yKesamaru

License

MIT License

Face_mosaic - Mosaic blur processing is applied to multiple faces appearing in the video

Related tags

Overview

動機

方法

静止画の場合

顔周りの長方形をぼかす

モザイク処理

動画に適用

結果

Reference

Author

License

Owner

Yoshitsugu Kesamaru

Python library to extract tabular data from images and scanned PDFs

Image processing using OpenCv

computer vision, image processing and machine learning on the web browser or node.

Fine tuning keras-ocr python package with custom synthetic dataset from scratch

This is a passport scanning web service to help you scan, identify and validate your passport created with a simple and flexible design and ready to be integrated right into your system!

FastOCR is a desktop application for OCR API.

One Metrics Library to Rule Them All!

Binarize document images

Satoshi is a discord bot template in python using discord.py that allow you to track some live crypto prices with your own discord bot.

When Age-Invariant Face Recognition Meets Face Age Synthesis: A Multi-Task Learning Framework (CVPR 2021 oral)

Responsive Doc. scanner using U^2-Net, Textcleaner and Tesseract

Markup for note taking

Slice a single image into multiple pieces and create a dataset from them

OCR, Scene-Text-Understanding, Text Recognition

Reference Code for AAAI-20 paper "Multi-Stage Self-Supervised Learning for Graph Convolutional Networks on Graphs with Few Labels"

CNN+LSTM+CTC based OCR implemented using tensorflow.

This repository contains the code for the paper "SCANimate: Weakly Supervised Learning of Skinned Clothed Avatar Networks"

This is the implementation of the paper "Gated Recurrent Convolution Neural Network for OCR"

Code for the paper "DewarpNet: Single-Image Document Unwarping With Stacked 3D and 2D Regression Networks" (ICCV '19)

This repository lets you train neural networks models for performing end-to-end full-page handwriting recognition using the Apache MXNet deep learning frameworks on the IAM Dataset.