Akinator Vision

• Download dataset - 3.8 MB
• Download full source code - 2.1 KB

Introduction

Akinator is a game that initially was created to predict famous characters by answering some questions to predict, but now Akinator wants to improve in Artificial Intelligence concept, especially Computer Vision. The new Akinator is working on Akinator Vision which recognizes famous characters only based off pictures and they want to add a new computer vision feature to some applications which have already been developed. This feature will allow the applications to detect and recognize every character based on profile image with a single face which has been uploaded to the developed applications. Therefore, as a programmer in Akinator Vision, you are asked to create a simple demo feature using Python programming language and OpenCv Library.

Details

1. Dataset Description

   The given dataset contains training dataset consisting of total 44 profile pictures of 7 users that already uploaded from the applications that was already developed by Akinator Vision and testing images consisting of 5 random user’s profile pictures.

2. Get Training Images Labels

   The directories of the given training dataset will be stored into a list containing the names of each famous characters. This list will also be used as the labels of the training images.

3. Get Training Images Data

   The image data will be stored into a list of images.

4. Detect Faces and Filter

   Faces inside the training images will be detected and stored into a list of images. The position and size of the detected faces will also be stored into a list of rectangles. You also need to filter the training images if there are no face or more than one face detected (Resize image width to 300 pixels and image height to the corresponding ratio for train images).

5. Training Data

   The list of face images which are already detected will be used to train face recognition classifier.

6. Get Testing Images Data

   The chosen testing images will be loaded and stored.

7. Predict Testing Images

   The list of testing images will be predicted to produce the prediction results based on the trained classifier above.

8. Write Prediction Results

   The prediction results which consist of the predicted names of the users will be drawn to every testing image. The rectangle around the face stored in the previous step will also be drawn together.

9. Combine and Show

   List of testing images that has been drawn will be combined onto a single image. After that, the program will show the combined result.

Using the Code

All the steps mentioned in the above have been put in the corresponding function in the code.

# Import Library Needed
import os
import cv2
import numpy as np

# Main
if __name__ == "__main__":
    '''
        Please modify train_path value according to the location of
        your data train root directory
        -------------------
        Modifiable
        -------------------
    '''
    train_path = "dataset/train"
    '''
        -------------------
        End of modifiable
        -------------------
    '''
    train_image_labels, train_image_indexes = get_all_train_image_labels(train_path)
    train_image_list = get_all_train_images(train_path)
    gray_train_image_list, _, gray_train_labels = detect_faces_and_filter\
                              (train_image_list, train_image_indexes)
    
    classifier = train(gray_train_image_list, gray_train_labels)
    '''
        Please modify test_path value according to the location of
        your data train root directory
        -------------------
        Modifiable
        -------------------
    '''
    test_path = "dataset/test"
    '''
        -------------------
        End of modifiable
        -------------------
    '''

    test_image_list = get_all_test_images(test_path)
    gray_test_image_list, gray_test_location, _ = detect_faces_and_filter(test_image_list)
    predict_results = predict(classifier, gray_test_image_list)
    predicted_test_image_list = write_prediction(predict_results, \
                                test_image_list, gray_test_location, train_image_labels)
    
    combine_and_show_result(predicted_test_image_list)

# Function to Labelling and Indexing Training Images
def get_all_train_image_labels(path):
    image_list = []
    image_index = []
  
    train_directory = os.listdir(path)
    for index, train in enumerate(train_directory):
        image_path_list = os.listdir(path + '/' + train)
        image_list.append(train)
        for image_path in image_path_list:
            if(image_path[-3:]=='jpg'):
                image_index.append(index)
    return image_list,image_index
    '''
        To get a list of path directories from root path
        Parameters
        ----------
        path : str
            Location of root directory
        Returns
        -------
        list
            List containing all train images label
        list
            List containing all train images indexes
    '''

# Function to Get All Training Images
def get_all_train_images(path):
    image_data = []
    train_directory = os.listdir(path)
    for train in train_directory:
        image_path_list = os.listdir(path + '/' + train)
        for image_path in image_path_list:
                if(image_path[-3:]=='jpg'):
                    image = cv2.imread(path +'/' + train + '/' + image_path)
                    image_data.append(image)
    return image_data
    '''
        Get all Train Images & resize it using the given path

        Parameters
        ----------
        path : str
            Location of root directory
        
        Returns
        -------
        list
            List containing all the resized train images
    '''

# Function to Detect Faces Using Haarcascade
def detect_faces_and_filter(image_list, image_labels=None):
    face_cascade = cv2.CascadeClassifier('haarcascade_frontalface_default.xml')
    image_location = []
    image_penampung = []
    label_penampung = []
    for idx,image in enumerate(image_list):
        faces = face_cascade.detectMultiScale(image, 1.2, 3)
        if len(faces)==1:
            for x,y,w,h in faces:
                image_location.append((x,y,w,h))
                image_crop = image[y:y+h,x:x+w]
                width = 300
                height = int(image_crop.shape[0] * width / image_crop.shape[1])
                dim = (width,height)
                image_crop = cv2.resize(image_crop,dim)
                image_crop = cv2.cvtColor(image_crop,cv2.COLOR_BGR2GRAY)
                image_penampung.append(image_crop)
                if image_labels is not None:
                    label_penampung.append(image_labels[idx])
    return image_penampung,image_location,label_penampung
    '''
        To detect a face from given image list and filter it if the face on
        the given image is not equals to one
        Parameters
        ----------
        image_list : list
            List containing all loaded images
        image_labels : list
            List containing all image classes labels
        Returns
        -------
        list
            List containing all filtered and cropped face images in grayscale
        list
            list containing image gray face location
        list
            List containing all filtered image classes label
    '''

# Function to do Training using LBPH Recognizer
def train(gray_image_list, gray_labels):
    lbph = cv2.face.LBPHFaceRecognizer_create()
    lbph.train(gray_image_list,np.array(gray_labels))
    return lbph
    '''
        To create and train face recognizer object
        Parameters
        ----------
        gray_image_list : list
            List containing all filtered and cropped face images in grayscale
        gray_labels : list
            List containing all filtered image classes label
        
        Returns
        -------
        object
            Classifier object after being trained with cropped face images
    '''

# Function to Get All Testing Images
def get_all_test_images(path):
    image_data = [] 
    image_path_list = os.listdir(path + '/')
    for image_path in image_path_list:
        if(image_path[-3:]=='jpg'):
            image = cv2.imread(path +'/'+ image_path)
            image_data.append(image)
    return image_data
    '''
        To load a list of test images from given path list
        Parameters
        ----------
        path : str
            Location of images root directory
        Returns
        -------
        list
            List containing all image in the test directories
    '''

# Function to Predict the Face Images in Test Set
def predict(classifier, gray_test_image_list):
    prediction_list = []
    for image in gray_test_image_list:
        lists, _ = classifier.predict(image)
        prediction_list.append(lists)
    return prediction_list
    '''
        To predict the test image with the classifier
        Parameters
        ----------
        classifier : object
            Classifier object after being trained with cropped face images
        gray_test_image_list : list
            List containing all filtered and cropped face images in grayscale
        Returns
        -------
        list
            List containing all prediction results from given test faces
    '''

# To draw prediction results on the given test images and resize the image Parameters
def write_prediction(predict_results, test_image_list, test_faces_rects, train_labels):
    train_labels = np.unique(train_labels) 
    images = []
    for idx,image in enumerate(test_image_list):
        x, y, w, h = test_faces_rects[idx]
        image = cv2.rectangle(image,(x,y),(x+w,y+h),(255,0,0),2)
        image = cv2.putText(image,train_labels[predict_results[idx]],\
                (x,y-10),cv2.FONT_HERSHEY_COMPLEX,1,(0,0,255),2)
        images.append(image)
    return images
    '''
        To draw prediction results on the given test images and resize the image
        Parameters
        ----------
        predict_results : list
            List containing all prediction results from given test faces
        test_image_list : list
            List containing all loaded test images
        test_faces_rects : list
            List containing all filtered faces location saved in rectangle
        train_names : list
            List containing the names of the train sub-directories
        Returns
        -------
        list
            List containing all test images after being drawn with
            final result
    '''

# To show the final image that already combine into one image
def combine_and_show_result(image_list):
    scale_percent = 60 # percent of original size
    width = int(image_list[0].shape[1] * scale_percent / 100)
    height = int(image_list[0].shape[0] * scale_percent / 100)
    dim = (width, height)
    image_list[0] = cv2.resize(image_list[0], dim, interpolation = cv2.INTER_AREA)
    width = int(image_list[1].shape[1] * scale_percent / 100)
    height = int(image_list[1].shape[0] * scale_percent / 100)
    dim = (width, height)
    image_list[1] = cv2.resize(image_list[1], dim, interpolation = cv2.INTER_AREA)
    width = int(image_list[2].shape[1] * scale_percent / 100)
    height = int(image_list[2].shape[0] * scale_percent / 100)
    dim = (width, height)
    image_list[2] = cv2.resize(image_list[2], dim, interpolation = cv2.INTER_AREA)
    width = int(image_list[3].shape[1] * scale_percent / 100)
    height = int(image_list[3].shape[0] * scale_percent / 100)
    dim = (width, height)
    image_list[3] = cv2.resize(image_list[3], dim, interpolation = cv2.INTER_AREA)
    width = int(image_list[4].shape[1] * scale_percent / 100)
    height = int(image_list[4].shape[0] * scale_percent / 100)
    dim = (width, height)
    image_list[4] = cv2.resize(image_list[4], dim, interpolation = cv2.INTER_AREA)
    combined_image = np.hstack(image_list)
    cv2.imshow("Final Result",combined_image)
    cv2.waitKey(0)
    cv2.destroyAllWindows()
    '''
        To show the final image that already combine into one image
        Parameters
        ----------
        image_list : nparray
            Array containing image data
    '''

Final Result

The dataset is obtained from Google Images. (Note: Some datasets are modified for the purpose of this case.)

History

• 24^th December, 2021: Initial version