Shulou(Shulou.com)06/01 Report--

How to use convolutional neural network and openCV to predict age and gender, this article introduces the corresponding analysis and solution in detail, hoping to help more partners who want to solve this problem to find a more simple and feasible method.

More and more applications are related to automatic classification of age and gender, especially since the rise of social platforms and social media. In spite of this, the performance of the existing methods on real images is still obviously inadequate, especially compared with the recently reported huge leap in the performance of face recognition-related tasks. -using convolutional neural networks for age and sex classification (https://talhassner.github.io/home/publication/2015_CVPR)

Introduction

Age and gender are two important attributes of human face, which play a very basic role in social communication, so estimating age and gender from a single face image has become an important task in intelligent applications, such as access control, human-computer interaction, law enforcement, marketing intelligence, visual surveillance and so on.

Real-world use cases:

I recently met Quividi, an artificial intelligence software application that detects the age and sex of users based on online face analysis and automatically starts advertising based on the target audience.

Another example might be AgeBot, an Android application that uses face recognition to determine your age from photos. It can guess your age and gender, and it can also find multiple faces in a photo and estimate the age of each face.

Inspired by the above use cases, we will build a simple age and gender detection model in this article. So let's start with our use case:

Use case-we will do some face recognition, face detection, and we will use CNN (convolution neural network) to predict age and sex from youtube videos, as long as the video URL is available, you do not need to download the video. The interesting part is that CNN is used to predict age and sex on the video URL.

Requirement: pip install OpenCV-python numpy pip install pafy pip install youtube_dl (learn more about youtube-dl: https://rg3.github.io/youtube-dl/)

The pafy:pafy library is used to retrieve YouTube content and metadata (such as title, rating, number of views, duration, rating, author, thumbnail, keywords, and so on). For more information about pafy, click on https://pypi.org/project/pafy/

Let's examine a sample:

Import pafyurl = 'https://www.youtube.com/watch?v=c07IsbSNqfI&feature=youtu.be'vPafy = pafy.new (url) print vPafy.titleprint vPafy.ratingprint vPafy.viewcountprint vPafy.authorprint vPafy.lengthprint vPafy.descriptionTesting file uploads with Postman (multipart/form-data) 4.8709678649911478Valentin Despa1688 ➡️? Check my online course on Postman. Get it for only $10 (limited supply): https://www.udemy.com/postman-the-complete-guide/?couponCode=YOUTUBE10I will show you how to debug an upload script and demonstrate it with a tool that can make requests encoded as "multipart/form-data" so that you can send also a file.After this, we will go even further and write tests and begin automating the process.Here is the Git repository containing the files used for this tutorial: https://github.com/vdespa/postman-testing-file-uploads

Steps to follow:

Get the video URL from YouTube.

Face Detection using Haar Cascade

Gender recognition of CNN

Age recognition of CNN

1. Obtain the video URL from YouTube:

Get the Youtube video URL and try to get the properties of the video using pafy, as described above.

two。 Use Haar cascaded face detection:

This is at least part of what most of us have heard of. OpenCV/JavaCV provides a direct way to import Haar cascades and use them to detect faces. I will not explain this part in depth. You can refer to my previous article to learn more about using OpenCV for face detection.

Article address: https://medium.com/analytics-vidhya/how-to-build-a-face-detection-model-in-python-8dc9cecadfe9

3. Gender identification of CNN:

Gender identification using OpenCV's fisherfaces is very popular, and some of you may have tried or read it. However, in this example, I will use different methods to identify gender. In 2015, two Israeli researchers, Gil Levi and Tal Hassner, introduced this method. I used the CNN model they trained in this example. We will use OpenCV's dnn package, which stands for "deep neural network".

In the dnn package, OpenCV provides a class called Net that can be used to populate neural networks. In addition, these packages support the import of neural network models from well-known deep learning frameworks such as caffe, tensorflow, and torch. The researchers I mentioned earlier have released their CNN model as a caffe model. Therefore, we will use CaffeImporter to import the model into our application.

4. The age recognition of CNN is similar to that of gender identification, except that the corresponding prototxt files and caffe model files are "deploy_agenet.prototxt" and "age_net.caffemodel". In addition, the output layer (probability layer) of CNN in the CNN consists of eight values from eight age groups ("0-2", "4-6", "8-13", "15-20", "25-32", "38-43", "48-53" and "60 -").

The Caffe model has two related files

1. Prototxt: here is the definition of CNN. This file defines each layer of the neural network, the inputs, outputs, and functions of each layer.

2. Caffemodel: contains information about training neural networks (training models).

Download .prtoxt and .caffemodel from here (https://talhassner.github.io/home/publication/2015_CVPR).

Download the haar cascade for face detection from here (https://github.com/opencv/opencv/blob/master/data/haarcascades/haarcascade_frontalface_default.xml).

Let's start coding our model.

Source code:

Import cv2import numpy as npimport pafy#url of the video to predict Age and genderurl = 'https://www.youtube.com/watch?v=c07IsbSNqfI&feature=youtu.be'vPafy = pafy.new (url) play = vPafy.getbest (preftype= "mp4") cap = cv2.VideoCapture (play.url) cap.set (3480) # set width of the framecap.set (4,640) # set height of the frameMODEL_MEAN_VALUES = (78.4263377603, 87.7689143744, 114.895847746) age_list = [' (0,2)','(4 6)','(8,12)','(15,20)','(25,32)','(38,43)','(48,53)','(60,100)'] gender_list = ['Male',' Female'] def load_caffe_models (): age_net = cv2.dnn.readNetFromCaffe ('deploy_age.prototxt',' age_net.caffemodel') gender_net = cv2.dnn.readNetFromCaffe ('deploy_gender.prototxt') 'gender_net.caffemodel') return (age_net, gender_net) def video_detector (age_net, gender_net): font = cv2.FONT_HERSHEY_SIMPLEXwhile True: ret, image = cap.read () face_cascade = cv2.CascadeClassifier (' haarcascade_frontalface_alt.xml') gray = cv2.cvtColor (image, cv2.COLOR_BGR2GRAY) faces = face_cascade.detectMultiScale (gray, 1.1 5) if (len (faces) > 0): print ("Found {} faces" .format (str (len (faces) for (x, y, w, h) in faces: cv2.rectangle (image, (x, y), (x, y), (255,255,0), 2) # Get Face face_img = image [y:y+h, h:h+w] .copy () blob = cv2.dnn.blobFromImage (face_img, 1,227,227), MODEL_MEAN_VALUES SwapRB=False) # Predict Gender gender_net.setInput (blob) gender_preds = gender_net.forward () gender = gender_ list [gender _ preds [0] .argmax ()] print ("Gender:" + gender) # Predict Age age_net.setInput (blob) age_preds = age_net.forward () age = age_ list [age _ preds [0] .argmax ()] print ("Age Range:" + age) overlay_text = "s% s"% (gender) Age) cv2.putText (image, overlay_text, (x, y), font, 1, (255,255,255), 2, cv2.LINE_AA) cv2.imshow ('frame', image) # 0xFF is a hexadecimal constant which is 11111111 in binary.if cv2.waitKey (1) & 0xFF = = ord (' q'): breakif _ name__ = = "_ main__": age_net, gender_net = load_caffe_models () video_detector (age_net, gender_net)

Now let's understand the code together:

Step 1: import all required libraries.

Import cv2import numpy as npimport pafy

Step 2: get the Youtube video URL and create an object "play" that contains the optimal resolution of the video in webm/mp4 format.

Url = 'https://www.youtube.com/watch?v=c07IsbSNqfI&feature=youtu.be'vPafy = pafy.new (url) play = vPafy.getbest (preftype= "mp4")

Step 3: usually, we have to capture the video stream of the scene with a camera. OpenCV provides a very simple interface. We can capture video from the camera, convert it into grayscale video and display it. It's just a simple start.

To capture video, you need to create a video capture object. Its parameters can be the device index or the name of the video file. The device index is just a number that specifies which camera. A camera is usually connected (as in my case). So I only pass 0 (or-1). You can select the second camera by passing 1 and so on. After that, you can capture it frame by frame.

Cap = cv2.VideoCapture (0) # if you are using webcam

But in my case, I am reading an online video URL, for which I will pass the "play" object to VideoCapture ().

Cap = cv2.VideoCapture (play.url)

Step 4: use set () to set the height and width of the video frame. Cap.set (propId, value), where 3 is the width of propertyId,4 is the height of propertyId.

Cap.set (3480) # set width of the framecap.set (4640) # set height of the frame

Step 5: create 3 separate lists to store Model_Mean_ values, age, and gender.

MODEL_MEAN_VALUES = (78.4263377603, 87.7689143744, 114.895847746) age_list = ['(0,2)','(4,6)','(8,12)','(15,20)','(25,32)','(38,43)','(48,53)','(60,100)'] gender_list = ['Male',' Female']

Step 6: I defined a function to load the age and sex detectors for caffemodel and prototxt, which are basically pre-trained CNN models for detection.

Def load_caffe_models (): age_net = cv2.dnn.readNetFromCaffe ('deploy_age.prototxt',' age_net.caffemodel') gender_net = cv2.dnn.readNetFromCaffe ('deploy_gender.prototxt',' gender_net.caffemodel') return (age_net, gender_net)

Step 7: now we will perform face detection, age detection, and gender detection, and create a function video_detector (age_net,gender_net) in your main function for this purpose, and take age_net and gender_net as its parameters.

If _ _ name__ = "_ _ main__": age_net, gender_net = load_caffe_models () video_detector (age_net, gender_net)

Step 8: read the cap object created from VideoCapture () in step 3. Cap.read () returns a Boolean value (True / False). If the frame is read correctly, it will be True.

So you can check the end of the video by checking the return value.

Sometimes, cap may not have initialized the capture. In this case, this code displays an error.

You can check whether it is initialized by the cap.isOpened () method. If it's true, go on. Otherwise, use cap.open () to open it.

Ret, image = cap.read ()

Step 9: convert the image to a grayscale image because the OpenCV face detector requires a grayscale image.

Gray = cv2.cvtColor (image, cv2.COLOR_BGR2GRAY)

Step 10: load the pre-built model for face detection.

Face_cascade = cv2.CascadeClassifier ('haarcascade_frontalface_alt.xml')

Step 11: now, how do we use cascade classifiers to detect faces from images?

OpenCV's CascadedClassifier makes it simple again, and detectMultiScale () can detect exactly what you need.

DetectMultiScale (image, scaleFactor, minNeighbors)

Here are the parameters that should be passed to detectMultiScale ().

This is a general function for detecting objects, and in this case, it will detect faces because we called this function in the face cascade. If a face is found, a list of the positions of the face is returned in the format of "Rect (XMagneyDifferent wdirection h)", and if not, "None" is returned.

Image: the first input is a grayscale image.

ScaleFactor: this function compensates for the size misperception that occurs when one face looks bigger than another, because it is closer to the camera.

MinNeighbors: a detection algorithm that uses a moving window to detect objects by defining how many objects are found near the current window before declaring the found face.

Faces = face_cascade.detectMultiScale (gray, 1.1,5)

Step 12: cycle through the face list and draw a rectangle on the face of the person in the video. Here, we basically look for faces, decompose faces, their size, and draw rectangles.

For (x, y, w, h) in faces: cv2.rectangle (image, (x, y), (x / w, y / h), (255,255,0), 2) # Get Face face_img = image [y:y+h, h:h+w] .copy ()

Step 13:OpenCV provides a function to help preprocess the image for deep learning classification: blobFromImage (). It executes:

Average subtraction scaling and optional channel swapping

So the blob of the blobFromImage4 dimension is created from the image. Optionally resize and crop the image from the center, subtract the average, scale the value by scale factor, and swap the blue and red channels

Blob = cv2.dnn.blobFromImage (image, scalefactor=1.0, size, mean, swapRB=True)

Image: this is the input image. We have to preprocess it first and then classify it through our depth neural network.

Scale factor: after we perform average subtraction, we can choose to scale the image by a certain factor. This value defaults to 1.0 (that is, no scaling), but we can also provide another value. Also note that the scale factor should be 1 / σ, because we are actually multiplying the input channel (after the average minus) by the scale factor.

Size: here we provide the amount of space expected by the convolution neural network. For most of the most advanced neural networks today, this may be 224 × 224, 227 × 227 or 299 × 299.

Mean: these are our average subtractions. They can be triples of the RGB method or a single value, in which case the supplied value is subtracted from each channel of the image. If you want to perform average subtraction, be sure to provide 3 tuples in the order (Rmag _ B), especially when using the default behavior of swapRB=True.

SwapRB:OpenCV assumes that the image is in BGR channel order; however, the average assumes that we are using RGB order. To resolve this difference, we can swap the R and B channels in the image by setting this value to True. By default, OpenCV performs this channel exchange for us.

Blob = cv2.dnn.blobFromImage (face_img, 1,227,227), MODEL_MEAN_VALUES, swapRB=False)

Step 14: predict gender.

# Predict Gendergender_net.setInput (blob) gender_preds = gender_net.forward () gender = gender_ list [gender _ preds [0] .argmax ()]

Step 15: predict age.

# Predict Ageage_net.setInput (blob) age_preds = age_net.forward () age = age_ list [age _ preds [0] .argmax ()]

Step 16: now we must use openCV's put text () module to put the text on the output frame.

The parameters of putText () are as follows:

Text data to be written

Place the location coordinates (that is, the lower-left corner of the start of the data).

Font type (check the cv2.putText () document for supported fonts)

Font scale (specify font size)

Conventional things, such as color, thickness, linetype, etc. For a better look, linetype = cv2.LINE_AA is recommended.

Overlay_text = "% s% s"% (gender, age) cv2.putText (image, overlay_text, (x, y), font, 1, (255,255,255), 2, cv2.LINE_AA)

Step 17: finally print your final output.

Cv2.imshow ('frame', image)

Finally, we have:

If cv2.waitKey (1) & 0xFF = = ord ('q'): break

Our program waits for the user to press a key for up to 1 millisecond. It then takes the value of the read key and compares it with 0xFF, which removes anything above the bottom 8 bits and compares the result with the ASCII code of the letter Q, which means that the user has decided to exit by pressing the Q key on the keyboard.

Output: video URL-1: https://www.youtube.com/watch?v=iH1ZJVqJO3Y

Video URL-2: https://www.youtube.com/watch?v=qLNhVC296YI

It's interesting, isn't it? But not very accurate.

This is the answer to the question about how to use convolution neural network and openCV to predict age and gender. I hope the above content can be of some help to you. If you still have a lot of doubts to be solved, you can follow the industry information channel for more related knowledge.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.